Open Reviews of the Proof of The Riemann Hypothesis

Of Kumar Eswaran: An Expert Committee’s Report

By
Dr. P. Narasimha Reddy (Editor)

June 2021

Sreenidhi Institute of Science and Technology

Yamnampet, Ghatkesar, Hyderabad 501301
 Open Reviews of the Proof of The Riemann Hypothesis
Of Kumar Eswaran: An Expert Committee’s Report
By
Dr. P. Narasimha Reddy (Editor)
This is a Report on the Open Reviews of the Proof of The Riemann Hypothesis of
Kumar Eswaran by the Expert Committee comprising of the following Members:

Members of the committee

1. Dr.T. Ramasami, Ph.D. (Leeds) – Honorable Former

Secretary, Ministry of Science and Technology,
Govt. of India - Advisor

2. Prof. P. Narasimha Reddy, Ph.D. (Kakatiya. Univ.) &

M.E. (I.I.Sc. Bangalore): Executive Director,
Sreenidhi Institute of Science and Technology - Chairman and Editor

3. Prof M. Seetharaman, Ph.D. (Univ. of Madras),

Formerly Professor and Chair Department of
Theoretical Physics, University of Madras - Member

4. Prof. V. Srinivasan, Ph.D. (Univ. of Wisconsin),

Formerly Professor and Dean School of Physics,
Hyderabad Central University - Member

5. Prof K. Srinivasa Rao, Ph.D. (Univ. of Madras)

Formerly Senior Professor, Institute of
Mathematical Sciences, Chennai - Member

6. Prof M. D. Srinivas, Ph.D. (Univ. of Rochester);

Senior Fellow, Centre for Policy Studies,
Formerly Professor of Theoretical Physics
Univ. of Madras - Member

7. Prof. Vinayak Eswaran, Ph.D. (Stony Brooke,

State Univ. of NY), Dept. of Mechanical
and Aerospace Engineering, IIT Hyderabad - Invited Member

8. Dr. Adindla Suma, Ph. D (Univ. of Leeds)

Associate Professor, Dept. of Computer Science,
Sreenidhi Inst. of Science and Tech. - Convener
 CONTENTS

Foreword pp i
Preface pp ii
1. Summary of Review and Comments of the Expert Committee pp 1-2
2. Letter of Invitation to Review RH by Dr P. Narasimha Reddy pp 3-5.

Part A : Report and Reviews of K Eswaran

3. Review of RH by Professor Seetharaman pp 6-7
4. Review of RH by Professor Vinayak Eswaran pp 8-35
5. Response to Review of Prof Vinayak Eswaran by K. Eswaran pp 36
6. Review of RH by Professor V. Srinivasan pp 37-39
7. Summary of Response by K. Eswaran on some Reviews pp 40- 42
8. Response to Review by Profs. Ken Roberts & SR Valluri pp 43-57
9. Email to Convener Dr. Suma, from Ken Roberts pp 58
10. Extract from Reprint of Paper by Prof S. Chandrasekar pp 59 -64
11.Review by Prof Ken Roberts and Prof SR Valluri pp 65-85
12.Review and Complete and Detailed Technical Discussions
with Prof W. Narkiewicz pp 86-140
13.Review of RH by Prof G. Sierra pp 141-143
14.Response to Review of Prof German Sierra by K. Eswaran pp 144-154

Part B: K Eswaran All Main Papers on RH

15.The Final and Exhaustive Proof of the Riemann Hypothesis
from First Principles pp 155 -187
16.The Pathway to the Riemann Hypothesis pp 188 -193
17.A Simple Proof That Even and Odd Numbers of Prime Factors
Occur with EqualProbabilities in the Factor-ization of Integers and
its Implications for the Riemann Hypothesis pp 194-199

NOTE: The earlier papers Notes of Dr Kumar Eswaran on RH can be found on his Project
page in Reserchgate: https://www.researchgate.net/project/The-Dirichlet-Series-for-the-
Liouville-Function-and-the-Riemann-Hypothesis
 FOREWORD
It gives me great pleasure to introduce the work of one of our Professors, Dr. Kumar
Eswaran, who had joined our Sreenidhi Group in 1999 almost from the time we had started
this Institution. He has been a Professor of Computer Science and has been teaching
various mathematical related subjects in Computer Science and Information Technology
ever since.

Almost 5 years ago he had found a proof of the famous unsolved problem namely “The
Riemann Hypothesis”. He had put up his proof on the Web and there were very many
downloads (numbering in several thousands), and given several lectures on his methods.
These lectures were well received and there were no unresolvable negative comments
however, in spite of all this there was a reluctance on the part of the Editors of International
Journals to put the paper though a detailed Peer review. Therefore after nearly three years
of this stalemated situation, we were advised by various scientists to form an Expert
Committee who can look into the proof of Dr. Kumar Eswaran.

So in January 2020 we formulated such an expert committee of Scientists. The Expert

Committee whose task was to evaluate Dr. Kumar Eswaran’s proof of the Riemann
Hypothesis was in a strange and difficult position. Normally, the correctness of a proof of a
mathematical problem of great importance is announced by scholars in mathematics who,
after having examined the proof carefully, willingly step forward to announce that the proof is
correct. But, despite the fact that Dr. Eswaran’s proof has been on the web for nearly four
years and has been sent to some mathematics journals, no one has seriously engaged with
it.

So the Expert Committee decided to invite over 1,200 experts (mathematicians and
theoretical physicists) across the world to participate in an open review of the proof of Dr.
Kumar Eswaran. The review was open in the sense that the referees had to be willing to
have their names and institutional affiliations openly revealed, so that nothing is done
anonymously, nothing can be said that would not be openly available for all other experts to
see. We could not think of a fairer way to get the proposed proof assessed.  But the status of
the Riemann Hypothesis in mathematics is so great that very few were willing to take the risk
of openly venturing an opinion on the correctness or the incorrectness of the proof. Only
seven scholars responded. The author of the proof responded to these comments, where
warranted. We are publishing the proof, the referees’ comments, and the author’s responses
in their entirety, without any redactions.

I can vouch for the fact that, under the circumstances, this expert committee has done due
diligence; it has done its utmost to conduct an intellectually rigorous, honest, and fair
assessment of the proposed proof. On the basis of the assessment, this expert
committee has concluded that Dr. Kumar Eswaran’s proof of the Riemann Hypothesis
is correct.

Dr. K.T. MAHHE

20th June 20, 2021
Chairman Sreenidhi Group of Institutions

Page i
 PREFACE (of E-Book)
During the month of January 2020, an Expert Committee was constituted by Sreenidhi Institute of
Science and Technology under the advice of Dr. T. Ramasami to examine the purported proof of the
Riemann Hypothesis by Dr. Kumar Eswaran. This step, as explained in an earlier public
communication by me, was necessitated by the fact that Dr. Eswaran’s purported proof titled “The
Final and Exhaustive Proof of the Riemann Hypothesis from First Principles", was lying extant on
the Internet for nearly four years and had received thousands of reads and downloads but had not
received the assent of a mathematical Journal to subject it to a detailed peer review.

We, as responsible scientists, believe that a paper on such an important problem as RH should not
suffer from a lack of review. All of us had listened to several presentations of Dr. Kumar Eswaran and
some of us have studied his work in depth. We believe that the proof merited a critical review by the
experts at the very least. Hence, we formed a self-assembled committee of mathematicians and
theoretical physicists and appealed to the world’s mathematicians and scientists to review the proof
by a transparent and open review system. It was intended that the names and institutional
affiliations of the reviewers along with their comments would be posted on a publicly accessible
website designed for this purpose. We believed that this open review being a transparent process
would deliver a fair assessment of the work and is probably the best method for evaluating a
putative proof of a problem of the intellectual challenge of RH. We invited mathematicians and
scientists of the world and requested them to review the proposed proof and offer their comments.
In this connection, more than a thousand invitation letters were sent worldwide by email in the
month of February 2020. In this preface I wish to report the results of the above open review, an
exercise which took 15 months.

I had written to more than 1200 eminent Mathematicians and Scientists in my capacity as Chairman
of the Expert Committee, inviting all of them to participate in a detailed and Open Review of the
work on RH by Dr.  Eswaran. However, though the downloads from the links provided were several
thousand, only a total of seven scientists accepted to participate in this open review and out of this
were two of our own Committee members who chose to take upon this onerous task and one of
them was by Invitation.

The Expert Committee have examined the reviews in detail and their comments are in their
Summary below. These are followed by the more detailed unexpurgated reviews of Dr. Eswaran’s
papers on RH, which are also compiled in this e-book.

Before closing, I would like to record my thanks to all the people who were involved in this evaluation
exercise: Firstly, the Reviewers: Prof. Ken Roberts, Prof. S.R.Valluri. Prof. WladislawNarkiewicz, Prof.
German Sierra, Prof. M.Seetharaman (Committee Member), Prof. V.Srinivasan (Committee Member)
and Prof. Vinayak Eswaran for the great deal of trouble they have taken to undertake their reviews.

My profound thanks to Dr. T. Ramasami (Adviser) and the other Committee members: Prof. K.
Srinivasa Rao, Prof. M.D. Srinivas. I also thank Dr. A Suma (the Convener) and the support staff
Ms. Padmini Suresh, Mr. Ankusham Satyanarayana and Mr. R. Dinesh.

Finally and most importantly we all thank Dr. K.T Mahhe, the Chairman of the Sree Group and the
management for their unstinted support for the conduction of this exercise.

Prof. P. Narasimha Reddy,

Chairman Expert Committee Page: ii
 SUMMARY OF REVIEWS AND COMMENTS OF THE EXPERT COMMITTEE

The Expert Committee have examined the Reviews of the purported proof of RH by Dr. Kumar
Eswaran, there were a total of seven reviewers but two of them have posted joint review these six
reviews are summarized using a few operational quotations from their reviews followed by their
overall comments.

REVIEW 1

By Prof. Ken Roberts and Prof. SR Valluri. Univ. of Western Ontario, Canada

In this joint review which extended by over 20 pages (pages 65-85) they said as follows:  they have
said as follows:

“We found Dr. Eswaran's work quite stimulating of mathematical ideas, and believe that his work
should be brought to the attention of a wider scholarly audience, That is, the proof (or selected
portions of the methodology) should be published. The Riemann Hypothesis has resisted the efforts
of many of the best mathematicians for over 120 years,...” (in email to convener page 58).
The committee has appreciated the painstaking work of the reviewers and records its thanks.

REVIEW 2

By  Prof. WladislawNarkiewicz, University of Worclaw, Poland, A well- known Polish Number

Theorist.

This review is in the form of very detailed technical discussions over emails conducted by Professor
WladislawNarkiewicz who had worked through many parts of the paper and asked queries and
examined the replies, the discussions extending nearly 60 pages (page 86 -140). The committee
commends the painstaking review and discussion which was conducted in the spirit of an open and
sincere investigation revealing the many subtleties of RH - we thank him. His entire discussion is
given in this report for the benefit of readers and posterity.

In his penultimate email Professor Narkiewicz said that arguments were “heuristic”, though he says
“I agree that the similarity of the considered sequence of values of the lambda-function with a
random walk gives some reasons to believe in the truth of the conjecture”(page 129)

Prof Narkiewicz’s final reply ended with this sentence: “I want also to stress that the word "heuristic"
has no negative meaning. A lot of work of really great mathematicians has been performed in a
heuristic way. This applies not only to old times (Euler, Laplace, the Bernoulli’s, ...) but also to recent
times”(page 139).

REVIEW 3

By Professor German Sierra, Dept. of Physics University of Madrid, Spain.

This is the only negative review (page 141-143) The Reviewer seems to have believed (erroneously)
that Eswaran was trying to prove the randomness of primes and also imputed that he (Eswaran) felt
Equal Probabilities is sufficient for the proof. Eswaran, in his reply,(page 144-154) protested that he
does not hold to either of these views. Since there was no reference by the Reviewer of more than
3/4th of the paper, Eswaran requested that the Reviewer kindly read the rest of the paper for the
details of the actual proof. Since the Reviewer did not respond, this Review has necessarily to be
treated as incomplete and infructuous.

Page 1
REVIEW 4

By Prof. M. Seetharaman, Formerly Dept. of Theoretical Physics Univ of Madras

This Reviewer after studying the papers of Kumar Eswaran, was very definitive and said the following;
“The author’s analysis is exhaustive, unambiguous, and every step in the analysis is explained in
great detail and established. The conclusions of the author and his result must therefore be
considered proven.” (Page 7)

REVIEW 5

By Professor V. Srinivasan, Formerly Professor of Physics and Dean Univ of Hyderabad

Professor Srinivasan reviewed the various steps of the proof saying that “by Judiciously using the
properties of the random walk problem”, it was shown “that Riemann’s Hypothesis is true. There is
also a numerical proof given.” “I compliment the author for solving the Riemann’s Hypothesis.”(Page
38-39)

REVIEW 6

By Professor Vinayak Eswaran, Dept of Mechanical and Aerospace Engineering, IIT Hyderabad.

Professor Vinayak, who is Kumar Eswaran’s younger brother had taken the trouble to spend the best
part of two years to understand and study the background material and understand the proof of RH.
Therefore he was invited by the Committee to write a review of the proof. He has submitted a very
detailed review that summarized all the arguments of the proof and says that there is no doubt that
the Riemann Hypothesis is proved. He also submitted an essay which discusses why the RH was so
difficult to prove, as there is perhaps only one way it could have been done (pages 8-35)

Readers not fully familiar with the details of the RH could first read DrVinayak’s review, as it gives
an excellent overview of Dr. Kumar’s proposed proof.

After careful perusal of all the arguments in the proof and the reviews from the experts who have
responded, the Committee felt that there are no negative arguments that could technically invalidate
the proof and therefore have arrived at the firm conclusion that the proof by Dr. K. Eswaran is both
credible and acceptable and that the RH can be considered as proven.

The Committee recommends that the detailed unexpurgated Reviews along with the comments of K.
Eswaran and his papers and the findings of this Expert Committee, be compiled in the form of an
e-book with a suitable Preface, be made available to the world’s scientific community for their
perusal and for the historical record.

Members of Technical Committee:

1) Professor M. Seetharaman, 2) Professor K. Srinivasa Rao, 3) Professor V.Srinivasan

4) Professor Vinayak Eswaran (Invited Member).

Page 2
Page 3
Page 4
 -3-

ON BEHALF OF COMMITTEE MEMBERS:

1) Prof K. Srinivasa Rao (formerly Senior Professor, Institute of Mathematical Sciences, Chennai
2) Prof M. Seetharaman, (formerly Professor and Chair Dept. of Theoretical Physics, Univ. of Madras)
3) Prof. V. Srinivasan, (formerly Professor and Dean School of Physics, Hyderabad Central Univ.)
4) Prof M.D. Srinivasan, (Fellow Institute of Policy Studies and Former Professor Univ. of Madras)
5) Prof P. Narasimha Reddy (Executive Director, Sreenidhi Institute of Science & Tech, JNTUniv. Hyderabad)
---
ADVISOR:
1) Dr. Thirumalachari Ramasamy , (Former Secretary to Government of India, Ministry of Science and Technology)
---
REFERENCES

The MAIN PAPER is contained in REF (1) below; however it is advisable to read Ref(2) first.

1) K. Eswaran: "The Final and Exhaustive Proof of the Riemann Hypothesis from First Principles
https://www.researchgate.net/publication/325035649_The_Final_and_Exhaustive_Proof_of_the_Riemann_Hypothesis_from_Fi
rst_Principles

2) K. Eswaran: A Pathway to the Riemann Hypothesis Preprint March 2019

https://www.researchgate.net/publication/331889126_The_Pathway_to_the_Riemann_Hypothesis

3) A Simple Proof That Even and Odd Numbers of Prime Factors Occur with Equal Probabilities in the Factorization of
Integershttps://www.researchgate.net/publication/324828748_A_Simple_Proof_That_Even_and_Odd_Numbers_of_Prime_Fact
ors_Occur_with_Equal_Probabilities_in_the_Factor-ization_of_Integers_and_its_Implications_for_the_Riemann_Hypothesis

4)Project Notes These contain work on the RH project, previous preprints, explanations and clarifications of the proof and
replies to questions and technical notes.
https://www.researchgate.net/project/The-Dirichlet-Series-for-the-Liouville-Function-and-the-Riemann-Hypothesis

5) The above website contains the PPT of several Invited lectures on RH. The Lecture delivered in IIT Madras is available here:
Invited Lecture on the Proof of RH IIT Madras, May 2019
https://www.researchgate.net/publication/333185552_Invited_Lecture_at_IIT_MADRAS_On_the_Pathway_to_the_Proof_of_th
e_Riemann_Hypothesis

LATEST DEVELOPMENT
Meanwhile, there has been a seven lecture series on Kumar Eswaran’s proof which was prepared by Prof Vinayak Eswaran ,
(Professor in IIT, Hyderabadand his brother), out of the latter’s own volition and interest. All Vinayak’s seven lectures which are at
the level of undergraduate STEM students are available in:
VINAYAK ESWARAN :Seven Lectures on Kumar Eswaran's Proof of the Riemann Hypothesis -Consolidated
https://www.researchgate.net/publication/336899740_7_Lectures_on_Kumar_Eswaran%27s_proof_of_Reimann_Hypothesis-
Consolidated
The Youtube version of his lectures are in: YouTube Lectures by Prof
Vinayakhttps://www.youtube.com/playlist?list=PLRsxymPrOKUAk3eXhK9FZdGAPmgGGrhfb&disable_poly
mer=true

Page 5
Part A : Report and Reviews on K Eswaran’s paper and his
replies
---------- Forwarded message ---------
From: seetharaman mahadeva rao <msr4777@gmail.com>
Date: Sat, Jan 2, 2021 at 11:18 AM
Subject: Report on Dr Kumar Eswaran's paper
To: NR RIEMANN <nrriemann@sreenidhi.edu.in>

Dear Dr. Suma,

Please find attached my report on Riemann Hypothesis by Dr. Kumar Eswaran. My best
wishes to him for getting due recognition for this important work.

Best regards,
M Seetharaman

RH_Report.pdf
69K

Page 6
 A report on the work of Dr. Kumar Eswaran on
Riemann Hypothesis (RH)

The author has tackled a very challenging problem (RH) that has eluded a fully satisfactory
solution thus far.

Instead of dealing with the zeros of the the zeta function ζ(s) directly, he chooses to consider
the inverse function F (s) = ζ(2s)/ζ(s) for complex s, which is analytic for R(s) > 1 and
whose poles exactly correspond to the non-trivial zeros of ζ(s). He then applies the powerful
complex variable theory and analytic continuation to explore the analytic properties of F (s)
to the left of the region R(s) < 1. Clearly RH is proved if it is shown that all poles of F (s)
lie on the line R(s) = 1/2. The paper has been devoted to essentially establishing this. The
crucial part of the analysis is in showing L(N ) < CN a+ for large N when 1/2 ≤ a < 1,
which will make F (s) analytic in the region a < R(s) and then actually determining that
a = 1/2. This establishes RH.

The author’s analysis is exhaustive, unambiguous, and every step in the analysis is explained
in great detail and established. The conclusions of the author and his result must therefore
be considered proven.

M Seetharaman
Professor (Retd.)
Department of Theoretical Physics
University of Madras

Page 7
 “The Final And Exhaustive Proof Of The Riemann Hypothesis From First
Principles”,
by Kumar Eswaran[1]
A Review and Comment1 by Vinayak Eswaran
Professor of Mechanical and Aerospace Engineering, IIT Hyderabad
Abstract
I believe that Kumar Eswaran has proved the Riemann Hypothesis. Supported at
one end by Littlewood’s RH Equivalent Statement (1912) and on the other by the
Law of the Iterated Logarithm of Khinchin (1924) and Kolmogorov (1929), the
proposed proof spans the distance between these two century-old classic results in
three steps, each of which can be understood with undergraduate mathematics, as I
show in this review.

Preamble
I am gratified to have received a request by the the Committee chaired by Dr P. Narasimha
Reddy and advised by Prof T. Ramasami2 to review Kumar Eswaran’s proposed proof
of the Riemann Hypothesis (RH). However, as I am an engineer by training, and am also
Kumar’s brother, I think some justification is necessary as to why I should be reviewing
a paper on the most famous problem in Pure Mathematics. So, in what follows below, I
will include an explanation as to why I feel qualified to comment on Kumar Eswaran’s
proposed Proof.

My own education (PhD in Mechanical Engineering, Stony Brook, 1986) and research
interest thereafter in computational fluid mechanics exposed me to a fairly high level
of mathematics of a purely practical engineering variety (Partial Differential Equations,
Linear Algebra, Probability theory, etc). But, until just a few years ago, the only thing
I knew of the Riemann Hypothesis was that it was a very important problem in Number
Theory that involved a lot of Complex Analysis, which too I gleaned from casual con-
versations with friends and colleagues who were mathematicians, rather than by direct
study. That changed one day in 2016 when Kumar told me that he had had a significant
break-through on the RH (which was what became the “Towers” proof of the Equal
Probabilities result I will discuss below). I read and understood the proof, perhaps not
quite fully appreciating it or its true significance, nor its context in what I had hitherto
believed was essentially a problem in Complex Analysis. I was told it was relevant to an
Equivalent Statement of the RH.

Still with no idea where it came from or how it was derived, I learnt that statement
was : The Riemann Hypothesis3 is equivalent to the condition that for every  > 0
ΣN
n=1 λ(n)
lim 1 = 0, >0 (1)
N →∞ N 2 +
1
The formal review is upto page 15, followed by my informal comment on the proof and its reception.
2
ex-Secretary for the Department of Science and Technology (DST) India (a position equivalent to
the Head of NSF in the USA). https://en.wikipedia.org/wiki/Thirumalachari_Ramasami
3
RH statement: The non-trivial zeros of the zeta function ζ(s) all lie on the Re(s) = 21 line.

Page 8
where the Liouville integer function is defined as λ(n) = (−1)Ω(n) and Ω(n) is the number
of prime factors, multiplicity included, in the integer n. Briefly, λ(n) is +1 if n has an
even number of prime factors, and −1 if it has an odd number (which includes primes),
and by definition λ(1) = 1. For convenience, wePwill use the Liouville series for the
summatory Liouville function, defined as L(N ) ≡ N n=1 λ(n)

Kumar’s interest in that Equivalence Statement was triggered by a comment he read4

in a monograph [6] that contained a compendium of known results on the Riemann Hy-
pothesis. Among them was the statement above which was verbally interpreted by the
authors as: “...the Riemann Hypothesis is equivalent to the statement that an integer
has an equal probability of having an odd number or an even number of prime factors”.
In an inspired hour, sometime between midnight and dawn, Kumar had come up with
the “Towers” proof of Equal Probabilities.

It soon became obvious that that mere proof of Equal Probabilities was not enough,
1
because of the denominator N 2 + appearing5 in the statement. However, the peculiar
form of that denominator, with its 21 exponent, and the numerator comprising a sum of
+1’s and −1’s reminded Kumar of an old paper [8] of S.Chandrasekar, the astrophysicist,
on random walks and random flights that he had read (as, it turned out, had I) and Ku-
mar had his second inspiration: If the Liouville series is a random walk, the RH is true.
At first blush, this looked impossible. The λ’s in the sequence are so rigidly deterministic,
prescribed by the formula λ(n) = (−1)Ω(n) , where Ω(n) is the number of prime factors in
n, that there is not the slightest hint suggestive of randomness in the series. Apart from
sheer formulaic resemblance, there is little to suggest that the Liouville series could be a
random walk.

However, the formal random walk requirements are merely these two (1) that the
probabilities of λ’s of ±1 be equal, and that (2) the consecutive λ’s be independent of
each other. The first seemed to be true, but the second seemed false for the Liouville
series. However, there was a possible escape from the latter conclusion: the truth or
falsity of the RH would be determined by the λ(n)’s as n → ∞. Was it possible that the
λ’s could become independent in that limit? But how can you show that?

The solution emerged when Kumar began investigating the origins of the Equivalent
Statement (1). It was founded on an analysis published by Littlewood[2] in 1912, and
reproduced in Edwards [4]. Littlewood started with the simple idea that if the zeta
function had a zero at any point s on the complex plane, i.e., ζ(s) = 0, then its reciprocal,
1/ζ(s), would have a pole at that point. By using the original infinite series definition of
the zeta function and the Euler product formula, he obtained 1/ζ(s) as
∞
1 X µ(n)
= , Re(s) > 1 (2)
ζ(s) n=1 ns
4
Throughout his life since I can remember, Kumar has read advanced mathematical tracts for plea-
sure, the way other people read novels.
5
For equal probabilities alone, that denominator would only need have been N , as we shall see later.

Page 9
where the µ(n) is called the M öbius function and is defined as:

1
 if n has an even number of prime factors without multiplicity
µ(n) = −1 if n has an odd number of prime factors without multiplicity

0 if n is neither, i.e., it has some prime factor(s) with multiplicity


It was known that the non-trivial zeros of the zeta function were in the “critical strip”,
0 < Re(s) < 1, as complex conjugate pairs symmetrically placed around the line
Re(s) = 21 . The equation (2) is valid only for Re(s) > 1 but Littlewood proposed to
analytically continue it into the critical strip. Of course, such continuation would fail at
a non-analytic singularity, like a pole, so if 1/ζ(s) can be continued though 12 < Re(s) ≤ 1,
proving there were no poles there [and, by symmetry, in 0 < Re(s) < 21 ], the Riemann
Hypothesis would be proved.

Littlewood did not prove the RH but did obtain the following Equivalent Statement
to the RH through his analysis:

ΣN
n=1 µ(n)
lim 1 = 0, >0 (3)
N →∞ N 2 +
where the M öbius integer function µ(n) is defined above. The similarity of (1) with (3)
is obvious. Indeed, the Equivalent Statement (1) is obtained in exactly the same way
as (3), by replacing 1/ζ(s) by the function F (s) ≡ ζ(2s)
ζ(s)
in Littlewood’s analysis. Like
1/ζ(s), F (s) too has a pole at every point in 0 < Re(s) < 1 where ζ(s) has a zero6 , and
if continuable through 21 < Re(s) < 1 would prove the RH.7

Kumar did not find the derivation of (1) in Edwards[4], and did it himself. First he
needed to obtain the canonical form of F (s) to be analytically continued by Littlewood’s
method: ∞
ζ(2s) X λ(n)
F (s) ≡ = , Re(s) > 1 (4)
ζ(s) n=1
ns
the similarity of which with (2) engenders the already-mentioned similarity of (1) with
(3). However, most unexpectedly, it was the derivation of (4) that contained the critical
clue that led to the (first) independence proof of the λ’s that Kumar was searching for,
which I will outline further on in this review.

Kumar then came up with another proof of the independence of the λ’s of the Liouville
series by showing that any dependence relationship of the type λn = f (λn−1 , λn−2 , λn−3 , ..., λn−M ),
is impossible for finite M (I will also summarise the argument later in this review). It was
also discovered that Littlewood’s result combined with the Prime Number Theorem gave
6
It has an extra pole at Re(s) = 12 , which is irrelevant to our purpose as it is outside the 21 < Re(s) < 1
region of continuation.
7
It is uncertain who first suggested F (s) be studied. But the derivation of its Equivalent statement
is exactly analogous to the derivation by Littlewood for 1/ζ(s), so the Equivalent statement (1) is here
attributed to Littlewood.

Page 10
an alternate proof of the Equal Probabilities result. So now he had two proofs each for
the Equal Probabilities and Independence results, and thus the L(N ) series was shown
to be a random walk as n → ∞.

It is easy to show [9] that the RH equivalent statement (1) is unaffected if any finite
number N0 of leading λs are dropped from the series, and n = 1, 2, ... is then counted
from the (N0 +1)th . This means that the RH is affected only by the behaviour of L(N ) as
n → ∞, where it has been proved to be a random walk. What does that mean for the RH?

The first
√ and most obvious result is that the summatory Liouville function L(N )
scaled by √N , becomes a standard Gaussian variable, in the limit of large N . That is,
ξ ≡ L(N )/ N will have a Probability Density Function P (ξ):
1 ξ2
P (ξ) = √ e− 2 (5)
2
that seems to probabilistically ‘ensure’ that ξ would have low numerical values. Thus, for
N → ∞, the value of L(N 1
)
= Nξ → 0 for any  > 0 thereby satisfying the RH Equiv-
N 2 +
alent Statement (1). However, without more precise information as to how (1) would
be satisfied, i.e., with precise bounds, this argument would remain essentially heuristic,
unsatisfying as a rigorous proof of the Riemann Hypothesis.

Kumar was alerted by a friend of mine, Dr S.V.Ramanan, a physicist-turned-biologist

currently residing in Chennai who had read an early draft of Kumar’s paper, that he
should look at the possible relevance of a great result in Probability Theory, but little-
known outside that field, called the Law of the Iterated Logarithm (LIL). Kumar did look
at it, and within a few hours realised that the LIL[10] put his proposed proof on a firm
basis, as a true proof8 , not just a heuristic one.

As Kumar did all this, I followed his every step, as we live in the same city. I could
follow much of Kumar’s reasoning, as I have a long acquaintance with random walk
models, albeit in a very different application. Bookended on one side by a long-accepted
Equivalent Statement that appears in standard texts ([4], [5], [6]) on the RH, and on the
other by a seminal result obtained by two great pioneers of modern probability theory,
there was little in Kumar’s own contribution that seemed to be out of the ken of an
undergraduate in Science or Engineering. In fact, to confirm to myself that Kumar’s
proof was tenable, I spent the better part of a full year to understand the mathematical
context of the Riemann Hypothesis and the arguments of Kumar’s proposed proof and,
to test my own understanding, wrote a 7-lecture exposition of the proof that would be
comprehended by an undergraduate. The lectures contain the full proof, including the
origin of and relevant historical developments on the RH. While it does not cover some
alternate proofs offered in Kumar’s original work, and the empirical statistical evidence
he gives on the “L(N )-series-is-a-random-walk” idea, I believe that my lectures [7] are
the most readable version of Kumar’s proof. I recommend them for your perusal!

8
I suppose there are mathematicians who would still argue whether a probabilistic proof is a true one,
but I will let them argue that matter with the ghost of Kolmogorov!
4

Page 11
 With this contextual background, let me now make an assertion that otherwise would
seem absurd for a person not an expert in Number Theory and Complex Analysis with
a particular specialisation in the RH to make: I fully believe Kumar Eswaran has proved
the Riemann Hypothesis. My only caveat would have been that the two pillars on which
the proof is constructed, namely, the Equivalent Statement of Littlewood and the “Law
of the Iterated Logarithm” of Khinchin and Kolmogorov, particularly the latter, are out-
side the zone of my competence. But these have both been accepted for over a century,
and their statements are easy enough to interpret and apply, and I am sure are applied
correctly in Kumar’s proof. Kumar’s part of the proof can be broken down into three
steps, all dealing with aspects of Random Walks, a subject I have had acquaintance with
for several decades. And each step has been proved in more than one way, always ac-
companied by transparent and compelling intuitional reasoning. As for the rest, there is
nothing in Kumar’s own contribution that is outside the scope of undergraduate mathe-
matics. Hence my confidence that Kumar’s proof of the Riemann Hypothesis is correct.

Outline and review of the proposed proof

I will devote the rest of my review to outline the various aspects of the proof, to show
that that they can indeed be understood by an undergraduate. As mentioned, there
are three steps of Kumar’s proof: The first two consist of showing that the summatory
Liouville function L(N ) is a random walk (of unit variance and “time-step”), as N → ∞,
by showing that (a) the probabilities are equal that the λ(n)’s equal ±1, and that (b)
λ(n) is independent of the λ’s of the numbers preceding n by any finite distance. The
third step (c) is to show that L(N ) being a random walk, as N → ∞, is sufficient to
prove the Equivalent Statement (1) and thus the Riemann Hypothesis. This requires the
application of the LIL of Khinchin and Kolmogorov. What follows now is largely taken
my Lectures [7], which can be referred for the details.

Proof of equal probabilities

The first step in Kumar Eswaran’s proposed
PN proof of the RH is that the λ’s of the
summatory Liouville function L(N ) ≡ n=1 λ(n) have equal probabilities of being ±1,
in the limit of N going to infinity.

1. The Towers proof

In his original paper [1] Kumar did this by the “Towers proof” wherein he reorganised
the natural numbers into a system of infinite subsets such that every natural number
(except 1) is systematically, i.e., algorithmically, placed in one and only one of these
subsets. These subsets are such that their members are ordered strictly by size, with
each member with λ = 1 being followed by another with λ = −1, and vice-versa. Such
a system is not unique, as Kumar later proposed another system [11].

Page 12
 Let us consider the system from the original paper. For every natural number n,
other than 1, we can do this: factorise the number and represent it as n = mP L , where
P is its largest prime factor, with multiplicity L, and m is product of all the factors
(including multiples) smaller than P (if any, otherwise m = 1). From the prime factori-
sation theorem it follows that this representation, with (m, P, L), is unique for each
n. We then place the number as the Lth member of the “Tower”, i.e., infinite subset
P
Sm = {mP 1 , mP 2 , mP 3 , .... mP L , ...}. As each consecutive member has one more prime
factor (as we include multiplicities) than the previous one, the λ values of the consecutive
P
members of Sm will alternate between λ = ±1. Its members are also ordered by size as
each member is larger than the previous one. It is clear that each natural number n > 1
P
will appear in one, and only one, of the subsets Sm , and that the system will represent
all the natural numbers, and each one only once, and so constitutes an alternate repre-
sentation of the natural number system (except n = 1).

The subsets each have a direct correspondence with, say, the natural sequence of
odd and even numbers in the number system in that every odd number (λ = −1, say)
is followed by an even number (λ = 1), and vice-versa, and are ordered strictly by
size. So just as we can say that a “randomly chosen” natural number would with equal
probability be even or odd, a randomly chosen member of any subset of this system will
equally probably have a λ of ±1. Now considering a “randomly chosen” natural number,
we can say that it will be placed in one unique subset of the system, where it will be
equally probable to have a λ of ±1. Thus a “randomly chosen” natural number9 will
be equally probable to have a λ of ±1. This, in essence, is Kumar’s Equal Probabilities
proof.10

2. Proof by the Prime Number Theorem

However elegant and interesting Kumar’s Equal Probabilities proof was in its use of only
elementary arithmetic, there is another way to obtain the same result by the way of the
Prime Number Theorem (PNT) and Littlewood’s result. The PNT was also motivated
by Riemann’s famous 1859 lecture to the Berlin Academy where he proposed the RH (see
[3] for the historical details). In that lecture, Riemann had advanced an exact formula
for the Prime Counting Function π(n) that determines the number of primes less than
or equal to n. The leading term in this exact formula (whichR was proved in 1895 by H.
n dx
von Mangoldt) is the Logarithmic Integral function Li(n) = 2 log(x) (which “improves”
n
Gauss’ estimate π(n) ∼ log(n) ). [The remaining “error term” involves the zeros of the
zeta function and gives the context the Riemann Hypothesis].

H. von Mangoldt showed that the PNT would follow directly from the result that all
9
An interesting aspect of Kumar’s proof is that he does not need to specify how such a “random”
natural number would be chosen. Such an algorithm may not even exist.
10
Kumar goes into some detail to show that towers with their first λ equal to ±1 would be equal in
number. I do not think that proof is required for the probabilistic interpretation offered here, as the
first λ being +1 or -1 is irrelevant to the fact that the λ’s in that tower will become equally probable as
its members increase to infinity — which is all that is required in the Equal Probabilities proof.

Page 13
the non-trivial zeros of the zeta function have real part less than 1. The Prime Number
Theorem, which is the formal result that π(n) ∼ Li(n) asymptotically, was proved inde-
pendently by Hadamard and Poussin in 1896 by showing there are no zeros of the zeta
function on the Re(s) = 1 line11 ([3] pp 155-6). Much later, in 1949, Alte Selberg gave
an elementary proof of the PNT, i.e., without using Complex Analysis.

The above may be used with Littlewood’s result to obtain another proof of Equal
Probabilities. Littlewood’s result (see Lecture 3 of my Lecture Notes) says that the
analytic continuation of F (s) backwards from Re(s) > 1 will be possible through a <
Re(s) ≤ 1 if and only if
L(N )
lim = 0,  > 0 (6)
N →∞ N a+

and that the poles of F (s) corresponding the non-trivial zeros of ζ(s) would then lie
between 1 − a ≤ Re(s) ≤ a by symmetry considerations.

The PNT shows there are no zeros of the zeta function on the Re(s) = 1 line, that is,
it must be the case that a < 1. This means that we can always find an  > 0 such that
a +  = 1. So the PNT implies that

L(N ) N+ − N− N+ N−
lim = = − = P + − P− = 0 (7)
N →∞ N N N N
where N + and N − are the numbers of +1’s and −1’s in the N λ’s in L(N ), so N + +N − =
N and L(N ) = N + − N − , and P+ and P− are the respective probabilities of λ = ±1, as
N → ∞. The PNT thus implies that the two probabilities are equal, as N → ∞, thereby
proving the Equal Probabilities result.

As (7) is an equivalent statement of the PNT, Kumar’s Towers proof of Equal Proba-
bilities is an alternate proof of the PNT. The ordering by size within the infinite subsets
used in that proof ensures that the members in each subset are sequentially included in
the L(N ) series as N increases, with the probabilities of λ = ±1 becoming more exactly
equal in each subset as more members are included, ensuring that, as N → ∞, P+ = P−
in each subset and as a whole in the entire number system. Whether Kumar’s proof will
constitute an elementary proof of the PNT, I will leave for experts to decide. For, while
the Towers argument uses only very elementary arithmetic, the connection to the PNT
is through Littlewood’s result, which does use Complex Analysis.

I have spent more space here on an alternate proof, which was not mentioned in Ku-
mar’s original papers, than on Kumar’s own proof of the Equal Probabilities result only
to make the point that the said result is secure for Kumar’s approach towards the RH.

11
which was all that was necessary, as it was already known that there could be no zeros with Re(s) > 1.

Page 14
Proof of the independence of the λ’s
The second step in Kumar Eswaran’s proposed proof of the RH is that the λ(n)’s of the
summatory Liouville function are independent, as n → ∞, of the λ’s of the numbers
preceding n by any finite distance.

The first proof

As mentioned above, the genesis of this proof came from the derivation of the canonical
series form of F (s) ≡ ζ(2s)
ζ(s)
that is analytically continued by Littlewood’s method to ob-
tain the Equivalent Statement (1).

In the region Re(s) > 1, the zeta function has two alternate forms:
∞
X 1 1
ζ(s) ≡ = Q 1 , Re(s) > 1 (8)
n=1
ns (1 − ps
)
p

where the second comes from Euler’s product formula and involves an infinite product
over all primes. The derivation of (4) requires us to start by using this product form
(1 − p1s )
Q
ζ(2s) p 1
F (s) ≡ = Q 1 =Q , Re(s) > 1 (9)
ζ(s) (1 − p2s ) (1 + p1s )
p p

and then express each term in the last product by the well-known infinite expansion12
which are then all “multiplied-out” to get the canonical form for F (s) as (4). [The entire
derivation is given in my 3rd Lecture on Kumar’s proposed proof [7]]. The critical step
comes when the powers of the primes are grouped and replaced by the unique natural
number that is their product

p1 × p2 × ... × pq × pm −→ n

where p1 , p2 , .. can be repetitions of the same prime (i.e., multiplicity is allowed). Every
natural number appears once and only once in the series, which can therefore be written
as an infinite series in n, as is done in (4). But this simple step contains a clue that is
of utmost importance to the independence question. If the coefficient in (4) of the term
1
p
involving the single prime p is λ(p) then the coefficient of n, λ(n), is simultaneously
obtained in this step as

λ(p1 ) × λ(p2 ) × ... × λ(pq ) × λ(pm ) −→ λ(n) (10)

from which we obtain the true13 multiplicative property of the λ’s as:

λ(p1 × p2 × ... × pq × pm ) = λ(p1 ) × λ(p2 ) × ... × λ(pq ) × λ(pm ) (11)

12 1
1+x= 1 − x + x2 − x3 ... |x| < 1
13
which refers both to the values of the λ’s as well as their positions along the natural number sequence

Page 15
 In the infinite expansion and multiplication to obtain the canonical form of F (s), it
turns out that the coefficient is −1 for every term p1 with a single prime p. That is, λ(n) =
−1, if n is a prime. Substituting this information into (11) gives us λ(n) = (−1)Ω(n) , as
it was first defined. But to mechanically make this substitution of λ(p) = −1 is to lose
all the insight contained in (11) which tells us not just the value of λ(n) but also where
this value will be placed, i.e., at location n = p1 × p2 × ... × pq × pm in the sequence
of the λ’s of natural numbers.14 This is obviously of significance when we recall that if
the Liouville sequence is to be a Random Walk the λ’s must be independent of those
preceding them in that sequence.

Equation (11) tells us the dependence relationship15 of the λ’s in the Liouville series,
i.e., the λ in the nth location of the series is dependent on the λ’s in the pth th
1 , p2 ..., and
pm locations (where the pi ’s are the factors of n) and on nothing else. This implies16
th

(i) that λ(n) is independent of the λ of any prime number that is not among its factors,
and further that (ii) λ’s of any two numbers m and n that are co-primes, i.e., without
common prime factors, will be independent, and also that (iii) the λ’s of two numbers
m and n that do share common prime factors but have a greatest common factor smaller
than either one,17 will also be independent.18

This says the only possible dependence between m and n (m < n) will occur when n
is a multiple of m, i.e, when n = m × q, where q is an integer factor. To be dependent
on m, as the smallest possible factor is 2, n needs to be at least twice as large as m.
This means λ(n) cannot be dependent on the λ of any number between n/2 and n, i.e.,
it is independent of the (almost) n/2 λ’s preceding it in the Liouville sequence. Then,
as n → ∞, λ(n) is independent of any finite number of preceding λ’s. This completes
Kumar’s first proof of independence of the λ’s of the Liouville series (as n → ∞).

14
A useful alternate reading would be: In the sequence of the λ’s of the natural numbers, the λ at
location n will be −1 if n is a prime, otherwise, its λ will be the product of the λ’s of the prime factors of
n. For this proof it is important to think of λ(n) as a discrete binary variable determined by its location
n along the sequence of λ’s on the natural number line.
15
The dependence of the λ’s is merely a reflection of the multiplicative dependence of the natural
numbers, all of which can be obtained by the multiplication of the primes – which therefore are the only
independent numbers, upon which the composite numbers are dependent.
16
Henceforth, λ(n), or λ of n, should be read as the λ at location n of the natural number sequence.
17
meaning, each has some unique prime factor(s) not appearing in the other
18
Let G be the greatest common factor of m and n (and smaller than both). From the multiplicative
property we get λ(m) = λ(G) × λ(P ) and λ(n) = λ(G) × λ(Q), where P and Q will be co-primes and
thus have independent λ’s, which will make λ(m) and λ(n) also independent.

Page 16
The second proof of independence
It is easy to show [9] that the Equivalent Statement (1) is unaffected if any finite number
N0 of leading λ’s are dropped from the series, and n = 1, 2, ... is counted from the
(N0 + 1)th natural number: That is,

ΣN
n=1 λ(n + N0 ) LN0 (N )
lim 1
+
= lim 1 = 0, >0 (12)
N →∞ (N ) 2 N →∞ (N ) 2 +

constitutes another equivalent statement of the RH, with N0 being any finite number. So
any number of leading λ’s can be dropped from the Liouville series and (12) would say
the same thing as (1) regarding the truth or falsity of the RH, both being RH Equivalent
Statements. This means that the truth or falsity of the RH is determined entirely by the
behaviour of the λ’s “at” infinity19 . So, the λ(n)’s that concern us are those at n → ∞.

Further, if, say, N0 = 10100 , it is evident that most of the prime factors that directly
affect the λ’s of the “shifted” series LN0 (N ) (≡ ΣN n=1 λ(n + N0 )) will not appear in the
series themselves.20 So there will not be the direct influence on the λ(n)’s by the con-
stituent primes of n, as shown in (11). However, there could still be the influence of
the λ’s of non-prime factors of n, i.e., products of some but not all of the prime factors,
which could create dependence relations in the LN0 series. So the dependency of λ(n)
could extend to its composite factors closer in magnitude to n. We now pursue a proof21
by contradiction, and first assume that such a dependency does exist, so that the LN0
series are not Random Walks.

Equation (12) actually embodies an infinite number of RH Equivalent statements, for

different N0 ’s, which must all be true if the RH is true and all false if the RH is false. So if
any LN0 series is not a Random Walk due to λ dependencies, they all should have similar
dependencies.22 However, the fact that N0 , in (12), could be any number means that the
sequence number n of the λ’s in the shifted series LN0 (N ) is effectively decoupled from the
natural number n + N0 that actually determines the λ values. Thus any dependencies
that exist between the λ’s, as n → ∞, would need to be recursive, i.e., refer to the
“local” n and its predecessors rather than the “global” natural number n + N0 . Further,
as n → ∞, any λ dependencies would reach an asymptotic and unchanging functional
form, independent of n, connecting the λ’s, as seen in (13) below. Kumar thus argues
that the only possible dependence relationship would need to be of the form:

λn = f (λn−1 , λn−2 , λn−3 , ..., λn−M ) (13)

and be operational as n and n + N0 go to infinity. If there is any such relationship, with

finite M , then the Liouville series would not be a Random Walk, as the independence
condition of the λ’s would be violated.
19
that is, in the endless expanse of numbers extending beyond all reckoning
20
as the smallest primes appear most frequently as factors of the natural numbers, with 2 appearing in
every second number and 3 in every third, while very large primes will be correspondingly less frequent
21
discussed in more detail in my Lecture 6 [7]
22
“at” infinity, which alone determines the RH

10

Page 17
 If the dependence relationship (13) is true then, as n increases toward infinity, each new
λ(n) in the Liouville series is determined by the previous M λ-values. However, we know
that, as the λ’s are binary-valued, the argument set of λn−1 , λn−2 , λn−3 , ..., λn−M could
have at most 2M different combinations. So, as n increases, the λ combination in the
argument of the dependence function in (12) would necessarily repeat some previously-
occurring combination, and yield the same λ(n) as before — and then the same λ(n + 1),
λ(n + 2)..., as before, so the series becomes cyclic thereafter. Further, the Equal Proba-
bilities result would ensure that each cycle would necessarily comprise equal numbers of
λ = ±1, so the L(N ) would return to the same value at the end of each cycle. Therefore,
L(N ) would remain finite even while N → ∞. If this happens, Littlewood’s condition
(6) would be satisfied for a = 0, implying that there are no zeros of the zeta function
between 0 < Re(s) < 1 – which is untrue because we know there are zeros23 on the
Re(s) = 1/2 line. Hence, the only possible dependency function (13) previously allowed
is now shown to be impossible. Thus the independence of the λ’s over finite strips is
assured as n → ∞, completing Kumar’s second proof.

The empirical evidence that L(N ) is a Random Walk

It has been shown above that the Liouville series L(N ) should become a Random Walk
as N → ∞, by proving that λ(n) values are equally probably ±1, and that they are
independent of the preceding λ’s, and therefore behave like “coin-tosses”, at least for
high values of n.

That the L(N ) series is a Random walk is a proposition that can be empirically tested
by statistical means in an exercise to, first, verify if indeed the proposition seems true
and, second, to throw some light on what that condition “N → ∞” means in practical
terms. The λ(n)’s of the series themselves can be obtained by brute-force factorisation
of n by mathematical software, and then using λ(n) = (−1)Ω(n) . Kumar did this in his
paper, and the results are interesting.

A standard and widely used test of statistical hypotheses is the Chi-squared (χ2 ) test
which comprises in computing a diagnostic χ2 value from the observed empirical data
and then checking whether that value falls within the normal range of values that would
occur if the null hypothesis were true. The “normal range” is left to the analyst’s choice
but usually includes 95% of possible χ2 values. If the empirical χ2 value falls outside this
chosen range, the test is said to have rejected the null hypothesis.

In our case we would like to see if the λ(n)’s of the L(N ) series from, say, n = n0 to
n0 + M − 1 are statistically distributed as if they were M coin tosses: this is the null
+ − )2
hypothesis. For this case χ2 = (M −M M
where M , M + and M − are, respectively, the
numbers of the λ’s in the stated range, and the numbers that are +1 and −1. The differ-
ence M + −M − can be directly obtained from the L(N ) series as L(n0 +M −1)−L(n0 −1).
+ −M − )
The difference of the probabilities of λ = ±1 in that range is ∆P ≡ P + −P − = (M M .
23
an infinite number, as proved by Hardy

11

Page 18
 2
For the full series (n0 = 1, M = N ), the χ2 value is χ2 = L(N
M
)
which is the square of the
L(N ) L(N )
proposed “normal gaussian variable” M , while ∆P = M . The expected χ2 distribu-
√

tion for this case is exactly Gaussian and χ2 ≤ 3.84 is the 95% bench mark for rejecting
the null hypotheses (if the computed χ2 value exceeds 3.84 more often than 5% of the
time).

Kumar did extensive statistical testing of the λ’s, over different and consecutive seg-
ments, typically with M =1000, and also the whole range, from n = 1 to 176 trillion,
which can be seen in the Appendix IV of his paper. He shows that over all segments
and the whole range the λ’s are statistically indistinguishable from coin-tosses in the
Chi-squared test, apart from showing ∆P steadily decreases to reach values below 10−7
by the end of the range. While it can be argued that n = 176, 000, 000, 000, 000, is not a
high number in the context of the RH, the statistical finding that L(N ) is indistinguish-
able from a Random Walk, right from the start of the series, gives powerful empirical
backing for the main line of attack of Kumar’s proposed proof.

Why a Random Walk L(N ) implies the RH is true

We have seen that, as N → ∞, the Liouville series L(N ) satisfy the two conditions24
required of a Random Walk, i.e., that (1) the probabilities are equal that the λ’s of
the series are ±1, and (2) that the λ(n)’s are independent of the λ’s of the numbers
preceding n by any finite distance. The remaining question, whether this proves the
Riemann Hypothesis, will be taken up in this section.

The Heuristic ‘proof ’

Liouville series L(N )25 is a Random Walk then, as previously mentioned,
If the the √
ξ ≡ L(N )/ N becomes a standard Gaussian variable as N → ∞, with zero mean and
unit standard deviation, which will remain probabilistically confined to low values (with,
say, the probability of |ξ| > 10 being around 10−23 ), thus ensuring that the value of
L(N ) ξ
1 + = N  → 0 for any  > 0 thereby satisfying the RH Equivalent Statement (1).
N2

This idea was the intuitional motivation for Kumar’s line of attack on the RH of
first showing that the L(N ) series is a random walk. However, having done that, the
argument put forward above would possibly have been at best considered heuristic, as
well as one that left important questions related to the RH (for example, the width of
the Critical Line) unanswered. A particularly niggling
√ question is that if we choose  = 0
in (1), Littlewood’s analysis predicts that L(N )/ N would be unbounded (as the zeros
of ζ(s) and the poles of F (s) would then necessarily be at Re(s) = 21 ) while the heuristic
√
argument would say that L(N )/ N would still be a Gaussian variable, and therefore
very much finite. However, all these questions were put to rest, and the proof was raised
24
That this has been shown only for N → ∞ does not matter in this context, as the Equivalent
Statement (1) is determined entirely by the behaviour of L(N ) as N goes to infinity.
25
or, more correctly LN0 (N ) shifted by a sufficiently high N0

12

Page 19
to a high level of rigour by the introduction of the Law of the Iterated Logarithm, as
shown below.

Formal proof through the Law of the Iterated Logarithm (LIL)

The LIL was respectively proved and reformulated in the 1920 by two of the great names
of modern probability theory, Khinchin and Kolmogorov. It looks at the maximum de-
viation that the Random Walk can make from the probabilistic zero-mean expectation,
during its transient evolution. Its importance in this context is that L(N ), assumed to
be a Random Walk, too would have deviations from the Random Walk zero-mean ex-
pectation, as n increases without limit. If the maximum deviation obeys a power-law of
N with an exponent greater than 12 , this would invalidate Equivalent Statement (1) and
disprove the Riemann Hypothesis. If, on the other hand, these deviations can be shown
to have no greater than an 12 exponent26 , as N → ∞, then (1) would be satisfied and the
Riemann Hypothesis proved. It is most fortunate that the LIL can be easily re-stated in
terms of (1).

The LIL27 , when applied to the Liouville series (now assumed to be a random walk
of zero mean and unit variance) gives the √ limit
√ superior (i.e., maximum value) of L(N )
1
28 log log logN
for N → ∞ as a.s. being Lmax (N ) = 2 N e 2 , where log log log N =
log(log(log(N ))) involves the nested (“iterated”) use of the natural logarithm.
Substituting this in (1) tells us:
√ 1 log log logN
√ 1 √
Lmax (N ) 2 e2 2 e 2 log log logN
2
lim 1 = = = lim = 0, >0
N →∞ N 2
+ N elogN N →∞ e(−dN ) logN
(14)
is necessary for the Equivalent Statement (1) to be satisfied, where dN ≡ log2log logN
logN
is a
monotonically decreasing function of N , with limit 0 as N → ∞. We see that no matter
how small  (> 0) is, dN will be smaller beyond some N , so that the last term goes to
zero, thus satisfying (14) and (1), so proving the Riemann Hypothesis!

Littlewood’s result (see Lecture 3 of my Lecture Notes) says that the analytic contin-
uation of F (s) backwards from Re(s) > 1 will be possible through a < Re(s) ≤ 1 if and
only if
L(N )
lim = 0,  > 0 (15)
N →∞ N a+
1
26
which is the lowest value it can have, as the mean magnitude of a Random Walk variable is O(N 2 )
27
Traditionally stated as [10]: Let {Yn } be independent, identically distributed random variables with
means zero and unit variances. Let Sn = Y1 + ... + Yn . Then
±Sn
lim sup √ = 1, a.s.
n→∞ 2n log log n

28
a.s.“almost surely”, with zero probability of violation. Because of symmetry, we do not need to
consider the limit inferior, the lowest value.

13

Page 20
and that the poles of F (s) corresponding the non-trivial zeros of ζ(s) would then lie
between 1 − a ≤ Re(s) ≤ a by symmetry considerations. As we have shown this for
a = 21 , the zeros will all lie on Re(s) = 21 , thereby also showing that the width of
the Critical Line is zero. By putting  = 0 in (14), we see that the LIL gives us
Lmax (N ) √
√
N
= 2 log logN a.s.. That is, the “Gaussian random variable” ξ will “almost
surely” go to infinity as N → ∞, which satisfyingly resolves the paradox√stated at the
end of the last section by agreeing with Littlewood that, indeed, L(N )/ N would be
unbounded as N → ∞ as the zeros of ζ(s) are all at Re(s) = 21 . Which concludes the
proof.

So there we have it. I have outlined the three steps of Kumar’s proposed proof of the
RH, starting from Littlewood’s Equivalent Statement (1) and ending with the the Law of
the Iterated Logarithm (LIL) of Khinchin and Kolmogorov. The first two steps involve
showing the λ’s of the Liouville series L(N ) appearing in (1) are equally probably ±1, and
are independent over finite strips, as N → ∞, thereby effectively becoming “coin-tosses”,
and making the L(N ) series a Random Walk. These steps have both been shown in two
different ways, where the first proof in each case is accessible not just to undergraduates
but perhaps also to high-school students. The third step, that L(N ) being a Random
Walk proves the RH Equivalent Statement (1), is shown by the LIL which establishes
that Littlewood’s Equivalent Statement (1) will be satisfied as N → ∞. Littlewood’s
method and the LIL are well-established results that are a century old, while Kumar’s
contribution in connecting them can be understood by undergraduate students. On these
facts I rest my conviction that Kumar has proved the Riemann Hypothesis.

***

References
[1] “The Final And Exhaustive Proof Of The Riemann Hypothesis From First
Principles”, by Kumar Eswaran. https://www.researchgate.net/publication/
325035649_The_Final_and_Exhaustive_Proof_of_the_Riemann_Hypothesis_
from_First_Principles 1, 5

[2] Littlewood, J.E., Comptes Rendus de l’Acad. des Sciences (Paris), 154, pp. 263-
266(1912). Translated in Edwards, H.M., (1974), Chapter 12 pp 260-263. 2

[3] Prime obsession - Bernard Riemann and the greatest unsolved problem in mathe-
matics, by John Derbyshire, Plume (2003). 6, 7, 21, 27

[4] Riemann’s zeta function by H.M. Edwards, Academic Press (1974). 2, 3, 4, 18

[5] The theory of the Riemann zeta function, by E.C.Titchmarch and D.R Heath-Brown,
Clarendon Press, Oxford (1986). 4

[6] The Riemann Hypothesis, by P.Borwein, S.Choi, B.Rooney and A.Weirathmueller

(Eds.) Springer (2006). 2, 4

14

Page 21
 [7] “Seven lectures on Kumar Eswaran’s Proposed Proof of the Riemann
Hypothesis”, by Vinayak Eswaran. https://www.researchgate.net/
publication/336899740_7_Lectures_on_Kumar_Eswaran%27s_proof_of_
Reimann_Hypothesis-Consolidated 4, 5, 8, 10

[8] Chandrasekhar S., ‘Stochastic Problems in Physics and Astronomy’, Rev.of Mod-
ern Phys. vol 15, no 1, pp1-87 (1943). Reprinted in Selected Papers on Noise and
Stochastic Processes, N.Wax (Ed), Dover Publications (2003). 2

[9] “The effect on the non-random-walk behavior of the Liouville Series by the first
finite number of terms”, by Kumar Eswaran https://www.researchgate.net/
publication/325390233_The_effect_of_of_the_non-random-walk_behavior_
of_the_Liouville_Series_LN_by_the_first_finite_number_of_terms?
enrichId=rgreq-2e61082c5b1d2134738e58abb72429ca-XXX&enrichSource=
Y292ZXJQYWdlOzMyNTM5MDIzMztBUzo2MzE4MTIyNTA0MTUxMTZAMTUyNzY0NzE4Nzc3OQ%
3D%3D&el=1_x_2&_esc=publicationCoverPdf 4, 10

[10] Law of the iterated logarithm, Wikipedia https://en.wikipedia.org/wiki/Law_

of_the_iterated_logarithm 4, 13

[11] Kumar Eswaran ResearchGate https://www.researchgate.net/publication/

324828748_A_Simple_Proof_That_Even_and_Odd_Numbers_of_Prime_Factors_
Occur_with_Equal_Probabilities_in_the_Factor-ization_of_Integers_and_
its_Implications_for_the_Riemann_Hypothesis 5

[12] Denjoy, A., (1931) L’ Hypothese de Riemann sur la distribution des zeros. C.R.
Acad. Sci. Paris 192, 656-658. 18

15

Page 22
 Some personal reflections on the proof and its reception
Having perforce lived with Kumar’s work these last several years, and having spent
much time to understand both his proposed proof and the context and history of the
Riemann Hypothesis, I have some personal thoughts that I would like to share, if I may,
both on the RH and on Kumar’s experiences on proving it. Yes, I shall assume, in what
follows, that Kumar has proved the RH – for it seems so obvious to me that he has – even
while I recognise that the jury to decide this, far from being out, has not yet even been
empaneled, well into the fifth year after Kumar first communicated his proposed proof
to the leading journals of mathematics, none of which even bothered to formally review it.

However, I do not want to restrict myself to Kumar’s woes, but intend to use my
unique position as the first person29 , after Kumar, to know the proof of the RH, and
count coup, so to speak, on this famous and enduring problem. So I shall return to
Kumar’s difficulties after having first unburdened myself of my thoughts30 on the proof.

The first question that would strike the mind of someone who, like me, has accepted
that Kumar’s proof is indeed correct, is why would it take so long to be discovered? All
the mathematics required in the proof is at least a century old – a century in which the
Riemann Hypothesis was widely regarded as the greatest problem in mathematics, and
every talented pure mathematician would have paused, at least for a while, to consider
its proof. Some of the greatest mathematicians of the 20th Century did, as is well-known.
Why did none of them succeed, particularly if the proof is so simple?

There are several reasons that can be given but, to my mind, the main one is this:
There was, and is, only one way to prove the Riemann Hypothesis. To borrow imagery
from Tolkien, the only way to breach the defences of the Lonely Mountain, which oth-
erwise would resist a frontal assault by large armies, is through a single secret door.
That door is the coin-toss behaviour of the Liouville integer function, λ(n), and the only
existing key to that door is the Liouville series for the function F (s). But again, why
is that? Because, I suggest, the random-walk behaviour of the λ’s is the reason why the
non-trivial zeros of the zeta function fall on the Re(s) = 1/2 line.

To say that something is the reason for something else implies causality. Physics has
the so-called arrow of time for distinguishing cause from effect between concomitants.
Mathematics has no comparable touchstone, and works through non-temporal logical
implications and equivalences alone to create structures of truths consistent with its ax-
ioms. So, in mathematics, to say that A causes B, needs an explanation. What does it
mean then to say that the random-walk of λ’s causes the zeros of the the zeta function
to lie on the Re(s) = 1/2 line?

29
Our brother Mukesh also read Kumar’s drafts but, living in the same city, I followed his work closely.
30
which have been greatly helped by my discussions with Kumar and Mukesh.

16

Page 23
 While mathematics lacks an arrow of time, it does have a hierarchal structure that
serves to establish precedence – the mathematics of various numbers, i.e., natural, inte-
ger, rational, irrational, real and complex numbers, are separate logical structures, but
each with axioms, elements, and operations that have been derived and generalised from
the preceding one, with the natural numbers being first. Each structure rests on the
previous ones like the successive floors of a building where the lower storeys were built
independently of the upper ones, but not the upper ones of the lower. So it is possible
to conceive of a mathematics of natural numbers without considering complex numbers,
but not a mathematics of complex numbers without considering natural numbers.

We have seen that the proof of the RH was arrived at by an Equivalent Statement
that tied the behaviour of the λ’s by an ironclad “if and only if” condition to the ze-
ros of the zeta function. The first is defined entirely by the natural numbers, while the
latter is embedded deeply in Complex Analysis. Equivalence is a two-way implication,
meaning that both propositions it contains are together true or together false. However,
it would be absurd to say that the behaviour of the λ’s is determined by the zeros of the
zeta function, but perfectly logical to say that the latter is determined by the former —
i.e., that the behaviour of the λ’s causes the zeros of the the zeta function to lie on the
Re(s) = 1/2 line, as no other condition is required for the latter.

However, even this does not entirely justify my further claim that there is only one
way to prove the Riemann Hypothesis. Consider then the prime numbers. Primes en-
ter the RH through the Euler product formula definition of the zeta function but are
“reabsorbed” into the natural numbers, i.e., lose their explicitly separate identity, in the
many equivalent representations of the zeta function. So whatever influence the primes
have on the zeta function is implicit, i.e., hidden. While the zeta function contains some
information of the primes – the zeros of the zeta function are used to make the Prime
Counting function more precise – it cannot also tell us why the zeros are located where
they are. That information has to come from elsewhere.

The first place to look, naturally, would be Complex Analysis or, stated more gener-
ally, Analysis. Could the peculiar behaviour of the λ’s have worked itself into the axioms
of Analysis, so that the RH could be proved entirely by Analysis? This seems impossible
because Analysis does not, fundamentally, even have the concept of prime numbers – as
every real or complex number is divisible – and so can say nothing of the λ’s, which are
entirely determined by the primes. This has a very significant implication — that the
Riemann Hypothesis cannot be proved by Analysis alone31 — which would explain why
it was so difficult to prove, as Complex Analysis was always the foremost line of attack.

It has sometimes been suggested, usually not very seriously, that the RH could possibly
be an example of the correct-but-unprovable-assertions of Godel’s First Incompleteness
31
Kumar informs me that some mathematicians have anticipated this, e.g., J.B. Conrey: “It is my
belief that RH is a genuinely arithmetic question that likely will not succumb to methods of analysis.”
https://www.ams.org/notices/200303/fea-conrey-web.pdf pg 353

17

Page 24
theorem, no natural examples of which have been found apart from the one constructed
by Godel himself. My argument here is that this is indeed true: The Riemann Hypoth-
esis, which is a statement in Complex Analysis, cannot be proved by Complex Analysis.
It required a bridge – provided by Littlewood’s analysis — out of Complex Analysis to
Elementary Number Theory, and is finally proved by Probability Theory!

Even granting that the location of the zeros of the zeta function are determined en-
tirely by the random-walk behaviour of the λ’s, could another way be found to prove the
RH other than by using the Equivalent Statement (1)? Is there another key to the secret
door into the RH? For this to be possible, whatever else that is used has to precisely
reflect the said behaviour of the λ’s. One candidate is the original Equivalent Statement
3 of Littlewood, involving the M öbius integer function µ(n) defined in (2).

It is easy to show that given the “coin-toss” behaviour of the λ’s, i.e, equal probabil-
ities and independence, that (3) indeed would be satisfied32 . This was actually done by
Denjoy [12] in the 1930’s. But he did not rigorously prove that µ(n) would have coin-toss
behaviour, as Kumar has done with the λ(n)’s, but instead gave plausibility arguments
for the equal probabilities and independence properties. It was not well-received. Ed-
wards ([4], pp 268-269) called Denjoy’s argument “quite absurd” and “ludicrous” – and
this in a technical monograph on the Riemann Hypothesis – which in retrospect we
see was somewhat unfair, as Denjoy had correctly identified the secret door to the RH,
even if he did not turn its key — which anyway, being (3), most likely was the wrong one.

While the proof of the Equivalent Statement (3), independently of (1), may be possi-
ble33 it is clear that the latter is the larger result, and subsumes the former. So I think
that I am right in saying that any proof of the RH would necessarily invoke the random-
walk behaviour of the λ’s – which is the only key to the secret door into the RH!

So the RH has its basis in that the λ’s behave like coin-tosses, and the summatory
Liouville function is a random walk. The sheer unexpectedness of this discovery — as
both seem to be perfectly deterministic and as far from random as could be possible —
no doubt is one reason why the RH-Equivalent Statement (1), appearing in every modern
32
The M öbius function µ(n) is equal to λ(n) for every n that is prime or a product of single prime
factors, and 0 if n has prime factors with multiplicity. So assuming coin-toss behaviour in the λ’s
(which would carry over to the non-zero µ’s due to the independence condition), the summatory M öbius
PN
function, M (N ) ≡ n=1 µ(n) will still be a random-walk, but for Np steps, not N , where Np (for large
N , π(N ) < Np < N ) is the total numbers up to N that have single prime factors. So we can say, as
with the Liouville function, that M (N ) will satisfy:

ΣN
n=1 µ(n)
lim 1 = 0, >0
N →∞ Np 2 +
N 1
and hence also (3) [by multiplying the above equation by ( Np ) 2 + ].
33
Kumar’s factorisation proof of independence of the λ’s will go through to the µ’s, but both equal
probabilities proofs for the λ’s have difficulties, possibly insuperable, of translation to the µ’s, as the
latter are non-zero only for a subset of the natural numbers, unlike the λ’s.

18

Page 25
textbook on the RH, was not recognised for what it was — the key to the puzzle of the
RH. And so it lay in plain sight, like the purloined letter in Poe’s story, to be discovered
by an amateur, while professional mathematicians hunted all over the higher domains of
mathematics, and even created new ones, for that elusive key.

The proof of the Riemann Hypothesis is, perhaps, the most anticipated event in math-
ematics. Number theorists tell us there are hundreds of research papers that assume
the RH is true to obtain results that would thereby become theorems the instant it is
proved. There are also dozens of equivalent statements, most involving very complex
mathematics, that again would be proven true along with the RH. Even without these
consequences, the RH itself is seen as the missing corner-stone of a beautiful edifice of
Analytical Number Theory that Riemann himself inaugurated. It is also expected that
the proof of such a long-standing problem would break fresh ground in mathematics and
bring in new ideas that would enliven future research. All this would be true with Ku-
mar’s proof.34

However a final expectation, that the proof would bring to bear new techniques, is
unmet. Let us consider this disappointment. The RH was proved by techniques known
for a century, a century in which mathematics made enormous strides. Divorced from
the RH, the key finding that the Liouville integer function is a random-walk would have
been seen as an interesting but inconsequential result obtained by old techniques, prob-
ably insufficient in itself even to guarantee tenure in a so-called great school.35 Yet, it
is most surprising that a proof that uses techniques known since the 1920’s defeated
Hardy, Littlewood, Hilbert, Selberg and Nash, to name just a few in a much longer list
of mathematicians who attempted it. Why?

I have already given the reasons. The RH, although formulated in Analysis, would
be impervious to Analysis for its resolution. Analysis would have, of course, been the
natural and primary mode of attack used by any 20th Century mathematician who at-
tempted a proof. But that would have been like the drunk’s searching under the street
light – because there was more light there – for the door key that was lost elsewhere.
That search was doomed to be fruitless.

However, even someone with an idea that the answer lay in Arithmetic, the true habi-
tat of prime numbers, would be lost searching for a needle in the proverbial hay-stack,
without even an idea what a needle looks like. Who would think that the behaviour of
the λ’s – entities that have significance only in Analysis, not Arithmetic – would provide
the ultimate resolution to the puzzle of the RH? It would have to be a fortuitous event,
34
Kumar informs me that the insights from his proof would be immediately applicable to certain
Dirichlet L-functions in the Generalised Riemann Hypothesis (GRH).
35
while, with the RH considered, it would inevitably confer on its discoverer – given, of course,
previous attendance and current employment in a great school and the right pedigree – the Abel Prize,
and if young enough, the Fields Medal and then the Abel Prize!

19

Page 26
somewhat like stumbling onto a secret door during a continent-wide search. But without
its key the significance of the door would never have been known. It is most fortunate
that a key to that door even existed, for — being forged by the unlikely-looking function
ζ(2s)
ζ(s)
— it quite easily may not have, making the RH forever out of reach of proof, driving
gifted mathematicians to insanity a thousand years from now. And the only hint of the
existence of that door and that key was the one-half exponent in the denominator of the
expression (1) that, like a sword stuck in an anvil, attracted the attention of a passerby
who intuited its meaning in an epiphany.36

So the proof has lessons, if not in techniques, for the broader endeavour of mathe-
matics. Riemann’s creation of Analytical Number Theory was hailed as a great advance
because it allowed the power and wider range of Analysis to be brought to the ancient
field of number theory. As mentioned, mathematics comprises several logical systems,
for the natural, integer, rational, irrational, real and complex numbers, stacked like the
storeys of a building, with each level gaining, through generalisation, greater power and
scope. So to bring the methods of complex and real analysis, from several levels above,
into play in number theory, the mathematics of natural numbers, was seen as an unqual-
ified good thing. What was, perhaps, not given as much thought is the possibility that
something could be lost in that process of generalisation — in this case, the immediacy
of the primes — so that the peculiar behaviour of the λ’s, which we have seen can be
understood by undergraduates, climbs up five levels and is so transformed in that process
that it becomes the greatest problem in mathematics.

However, to say that the path to the proof of the RH was a very narrow and improb-
able one does not imply that the result itself is inconsequential. The great beauty of the
Riemann Hypothesis lies in the connection that it makes between the natural numbers,
which surely must have been known to humans at least a hundred thousand years ago,
and the theory of complex variables, which is based on an unimaginable abstraction,
the square root of a negative number, that was discovered by the human imagination
less than three centuries ago. That connection is magnificent, like a bridge traversing
different worlds.

It is also a test of that bridge. An engineer, by attaching a sound source to one end
of an aircraft wing and a sensor to the other, can discern merely by using high-frequency
ultrasonic waves whether the wing is structurally sound. So too the Riemann Hypothesis
is a test of the entire structure of mathematics that had its origins in the natural number
system and reached an apogee, around the time of Riemann himself, in the theory of
complex variables.

36
If there is a divinity that presides over Mathematics, it surely must be a goddess – as Ramanujan
believed, and whom he credited for his inspirations. I like to imagine that the Goddess looked down from
the high heavens and saw a seventy-year-old man reading an abstract tract on the Riemann Hypothesis
late into a moonlit night, by the last light of Durin’s day, and said to herself, “Why, he loves me truly”,
and impetuously slipped him the intuition of the secret door into the Lonely Mountain, a boon she had
denied the greatest mathematicians of the 20th Century...
20

Page 27
 More than any other field of human endeavour, mathematics is a product of pure
thought, and its structural integrity lies entirely in the soundness of the logic underlying
that thought. As mathematics progresses by generalisation, its new fields are created by
adapting the operations used in the older ones to apply to newer conceptual creations. So
the ideas of addition, subtraction, multiplication and division, learnt on natural numbers,
are extended and generalised to fractions and negative numbers, and to the so-called real
numbers and then to complex numbers. The edifice is created layer by layer, with each
stage resting on the previous one. But the new discoveries and theorems almost always
pertain to the new field and rarely say anything novel and interesting about the older
ones. For example, non-Euclidean geometry, created in the 19th and early 20th Centuries,
added not a single new theorem to Euclidean geometry37 , created two millennia ago.

So the newer fields presume the truth of the older ones on which they are based, but
they do not discover new truths in those older fields. This leaves open a door to a chill-
ing doubt — is “higher” mathematics saying something different from the “lower” even
when it is speaking of the latter? The whole of mathematics was constructed so that this
should not be true. But could an insidious fallacy have crept in during its expansion and
generalisation, so that mathematics is no longer consistent?

In 1931, Gödel with his famous Incompleteness theorems, first raised the possibility
that axiomatic systems in mathematics — which could be the whole field or any of its
branches — may have some propositions that may not be either provable or disprov-
able. So mathematics is possibly incomplete. His second theorem went even further, and
said that no such logical system can ensure its own consistency — that is, guarantee
that its theorems would never contradict one another. Such a contradiction would, of
course, be disastrous and invalidate the entire system. Incompleteness one can perhaps
live with, but inconsistency is another matter — and far more serious. For it would mean
that mathematics of some level may say something wrong about another more basic or
foundational level. That, surely, is something that must keep the Spirit of Mathematics
awake on bad nights. Which is why even mathematics needs a system check as much as
an aircraft does. The Riemann hypothesis is just such a check.

But there have been others before. In 1734, Euler solved the so-called Basel prob-
lem (see, e.g., [3]), that had been proposed almost a century earlier, of theoretically
determining the sum of the infinite series38 :
∞
X 1 1 1 1
2
≡ 1 + 2 + 2 + ... + 2 + ...
n=1
n 2 3 n

that had resisted the attack of many well-known mathematicians of the day. Euler’s
2
answer — that the sum was π6 — was particularly stunning for it involved the transcen-
dental number π that had hitherto been known only as the ratio of the circumference of
a circle and its diameter, thus exposing an unexpected connection between the natural
37
although it did create a useful debate on the foundational axioms of Euclidean geometry
38
which turns out to be related to the zeta function — being precisely ζ(2)

21

Page 28
numbers and geometry, and seeming to hint that mathematics is less a human creation
than a human discovery. Most importantly, the result was verifiable, for the series con-
verges quite rapidly and the actual calculation of the first few dozen terms shows their
sum approaching ever closer to the value given by Euler for the infinite sum. This gave
confidence39 that the “higher” mathematics used to derive the result — already far re-
moved from the counting-of-natural-numbers origins of arithmetic — was sound.

The Riemann Hypothesis is the Basel problem on steroids, as they say. It involves
every significant step in mathematics leading from the natural numbers to what is called
Analysis — which had reached a very high level of achievement just around Riemann’s
era. To elaborate, the Riemann Hypothesis connects the entire sequence of mental leaps
from whole numbers, addition, subtraction, multiplication, division, fractions, negative
numbers, real numbers, imaginary numbers, real functions, infinite series, limits, conti-
nuity, differentiation, integration, complex function theory, analyticity, analytic continu-
ation, convergent and divergent series and then completes the circle by referring back to
the natural numbers. The end of all that exploration is to arrive where we started, as
the poet said.

So what finally matters with the Riemann Hypothesis may not be what it precisely says
about the natural number system but that is says something true — thereby validating
to some degree the structure of mathematical analysis that was created over hundreds
of years. This is important. For more than any other branch of mathematics, modern
science and technology depends on Analysis. Analysis allows aeroplanes to fly, radios
to hum, skyscrapers to stand, and E = mc2 to be written. It is no doubt this intuitive
understanding of its importance that drove Hilbert, Hardy, Littlewood, and scores of
other great analysts to give the Riemann Hypothesis its supreme place in mathematics
— its proof, to recall Kronecker’s remark, would justify the work of man before God, so
to speak. Therein may lie its true significance.

Having considered what the Riemann Hypothesis could mean to mathematics in gen-
eral, let us consider some of the specifics shown up by Kumar’s proof.

Fittingly, for a result that has obsessed number theorists for a century, the Riemann
Hypothesis includes the entire domain of natural numbers. Its truth depends on the
result of a head-count made by the summatory Liouville function, L(N ), comprising the
sum of the λ’s of every natural number, wherein the λ of 5, say, counts equally with that
of five thousand, or five trillion, or that of a 5 trillion digit number. Yet, to paraphrase
Kipling’s injunction, all numbers count with it but none too much: for, as we have seen,
we can drop any finite number of λ’s from the count, or even an infinite number — say,
that of every alternate number — and yet the truth of the Riemann Hypothesis would
39
As it turned out, prematurely, for part of the method used by Euler could be justified by mathemati-
cians only a century later. So the empirical verification turned out to give a premature commendation
of the logic used.

22

Page 29
be unwavering. The RH is truly the will of all natural numbers, freely and independently
exercised40 , and so the most democratic result in all of mathematics!

And it takes an extraordinarily large “poll” to even indicate the final result of that
collective will. We have seen, in equation (14), the function dN ≡ log2log logN
logN
represents
1
the deviation from the critical line, Re(s) = 2 , possible when the summatory Liouville
function is counted to some finite N . That is, it represents the “half-width” of the critical
line for finite N , going to 0 as N → ∞. Now dN , surely, is the slowest decreasing function
of any significance in mathematics. For N = 10100 , i.e. one googol, vastly more than all
the atoms in the Universe, dN = 3.7 × 10−3 . For N = 101000000 , dN = 5.7 × 10−7 , less
precision than that of a pocket-calculator. The RH truly spans infinity.

In the 20th Century, rising with digital computing, Monte Carlo techniques – numerical
simulations of stochastic processes ranging from solutions of Partial Differential Equa-
tions to variations of the Stock Market – required the generation of large numbers of
random numbers distributed in specific ways, i.e., uniformly between [0, 1], say, or as
standard Gaussian variables. Of these the [0,1] uniformly distributed random numbers
are key, as the others are obtained using them. Because they can only be generated
algorithmically, not “randomly”, they are called pseudo-random numbers. Sequences of
such numbers invariably betray their falsity by repeating their patterns after some cy-
cle length. The generation of pseudo-random numbers – because they would need to
satisfy the requirements of their specified probability distributions and mutual indepen-
dence, etc – is a sufficiently important problem that Knuth devoted to it 200 pages in
his encyclopaedic The Art of Computer Programming. Improvements in pseudo-random
generation have gone on for decades, as it is believed there is no such thing as a perfect
pseudo-random generator. Yet, we have found one — the Liouville series!

We have seen how the λ’s of the Liouville series, while being dependent upon previous
λ’s of their prime factors, quickly become independent of their immediate predecessors
so that the last n/2 of the natural sequence of any n λ’s are independent of each other,
causing an exponential growth of independence further along the series. Even the early
dependencies, for small n, are not detected by the χ2 test, as the numbers involved are
too few. So the Liouville series is effectively an infinite random sequence of +1’s and −1’s.

Consider the first few: +1, -1, -1, +1, -1, +1, -1, -1, +1, +1, -1, -1, -1, +1, +1,
+1, -1, -1, -1, -1, +1, +1, -1, +1, +1... Let us now convert this to a binary number
by interpreting the values to be the digits after a decimal point, with -1’s being made
0’s, as in: 0.1001010011000111000011011... which gives us an infinite [0,1] uniformly
distributed binary number — the first random number generated by the natural number
system!41 Further, we can chop up the Liouville series into blocks of M -length and obtain
an infinite supply of M -precision [0,1] uniformly distributed binary random numbers42 .
40
being, seemingly, coin-tosses but actually mostly determined by a structure of dependencies
41
It possibly deserves a bronze plaque, or something.
42
which can, of course, be directly converted to standard [0,1] random decimal numbers.

23

Page 30
By starting the M -blocks at differently chosen natural numbers P , we can get different
random number sequences.43 This creates a class of natural random number generators
which give uniformly distributed and never-repeating sequences of perfect random num-
bers of any chosen precision.

I could go on.44 But, not to further taint the purest of pure mathematics of the
Riemann Hypothesis by these atrocious practicalities – as Hardy would have protested –
let me now stop.

I have now come to the part that I have dreaded writing, although it first gave me
the motivation to attempt this essay. It is about the treatment that Kumar’s proposed
proof has received in the mathematics community.

I can understand how the first sentiment of any mathematician who has some ac-
quaintance with the RH on seeing Kumar’s proposed proof would be disbelief. How
could this problem that has defeated the greatest mathematical minds of the 20th Cen-
tury be solved by an amateur, and that too in such a simple and straight-forward manner,
using no mathematics less than a century old, bypassing without mention all the work,
after Littlewood and Hardy’s first excursions, on the Riemann Hypothesis – and the Gen-
eralised Riemann Hypothesis – by several generations of professional mathematicians?
It would, perhaps, constitute the greatest subversion of conventional expectations in the
history of modern mathematics. Nevertheless, I would say that professional integrity, if
not a more fundamental love for truth and beauty, demands that this work, accessible to
every mathematician, should be fairly and openly evaluated.

That has not happened. The proof was completed (except for the LIL part) very
rapidly in a few weeks around August-September 2016, and sent to the Proceedings of
the Royal Society – Series A shortly thereafter, which wrote back after an unexplained
delay of more than 8 months (July, 2017) that they would not formally review the pa-
per but helpfully guided Kumar to a website that gave advise to mathematicians who
thought they had solved a great problem! They also suggested that the paper be put up
43
However these sequences need not be unique and independent from each other. Very often, many
random number sequences are simultaneously generated and it is important that they not be correlated
with each other. Usually they would be of the same precision M (= 32, 64, 128,..bits, say), but would
have different starting P ’s. Even so, the ith 1 number of one sequence may fall on precisely the same
segment of the Liouville sequence as the ith2 number of another, after which they would become endlessly
the same. Such synchronicity, and other unintended correlations, can be avoided by skipping s + q
bits of the Liouville sequence after the ith number segment before determining the (i + 1)th number, in
each random number sequence, where s is the preassigned unique serial number of the random number
sequence (of the many being generated) and q is the (random) binary number indicated by the first Q
bits in the Liouville sequence after the ith number segment (e.g., for a prescribed Q=4 and corresponding
bits, say, 1010 ⇒ q=10).
44
For example, the Liouville infinite binary sequence could be the public key for an encryption al-
gorithm where the private key indicates the segments thereof which stand-in for the keyboard set of
alphanumeric characters, and so on.

24

Page 31
on a website so that other mathematicians could read and review it. Kumar had already
put up the draft on ArXiv.org (September 2016), but then put it on Researchgate along
with later updates (the final one in May 2018).

Kumar then send it to IHES Paris France: Publications Mathmatiques de l’IHS (Oct
2017), The National Academy of Sciences, USA (Jan 2018), International Mathematics
Research Notices (Feb 2018), and Ramanujan Journal (Aug, 2019). In all cases he re-
ceived a straight-forward refusal to review the paper, or very cursory review, no more
than a paragraph long, that dismissed the paper. Appeals for reconsideration were either
refused or just ignored. The reviews invariably gave the impression that the reviewer
had not read beyond a few pages, merely looking for a quick reason to reject the paper.

More than one editor suggested that Kumar ask a prominent number theorist to care-
fully read and recommend the paper for formal review. Kumar took up this suggestion.
Knowing no number theorists, or any eminent professional mathematicians, he sent the
paper with a cover letter to more than 30 mathematicians and theoretical physicists who
had published papers on the RH. Just two or three even replied to his letter, and none
agreed to review it. Not a single number theorist Kumar corresponded with, then or
subsequently, has pointed out a possible fatal error in the proof, or even acknowledged
reading the full paper. Kumar is stuck in the classic Catch-22 — the journals would not
formally review his paper till a number theorist had recommended it, and no number
theorist would even admit to reading the paper! In the meanwhile, the paper on the
Researchgate website, till the date of the this writing, has had several thousand down-
loads — many traceable to prominent academic institutions across the world — but not
a single public comment by any self-identified academic mathematician.

Contrast this to the way the world mathematics elite treats one of its own. Within
living memory, a longstanding great problem in number theory to be solved was Fermat’s
Last Theorem. Andrew Wiles, then a professor at Princeton University, proposed a proof
in 1993. It was immediately closely examined by other academic mathematicians, and a
flaw was found within a few months. Wiles, helped by another mathematician, took a
full year to discover a fix for the proof which was finally published in a dedicated issue
of the Annals of Mathematics in 1995, and Wiles became a star in the constellation of
mathematics genius. Wiles proof was 200 pages of very advanced mathematics. Kumar’s
proposed proof is comprehensible by an undergraduate, and yet, five years on, he still
awaits a sincere review of his paper by academic mathematicians!

Another great open problem in Pure Mathematics that was recently solved is the
Poincare Conjecture, that was proved by Grigori Perelman, a Russian mathematician,
around 2003. Perelman, an unconventional person in many ways, did not publish his
proof in academic journals but merely uploaded his papers to ArXiv.org and gave a
few talks on it at some universities. His work was taken up and its details completed
and explained by other mathematicians. Perelman, perhaps to express his disdain for
the mathematics establishment, refused all prizes for his work, including the Fields medal.

25

Page 32
 Western scientific thought, perhaps the greatest intellectual leap of humanity, spread
across the world in the 20th Century, which led to an apparent widening of its talent pool,
with names like Ramanujan, Raman and Chandrasekhar entering its pantheon. But this
democratisation is still largely illusory even today — for one would be hard-pressed to
think of a single Nobel Prize winner or a Fields medalist who was not working at some
great western university, let alone not educated at one. If genius is everywhere, it cer-
tainly is not evident by the geographical distribution of modern scientific achievements.

This has led to a wide prejudice, common in the West but prevalent even in the de-
veloping nations45 , that a breakthrough on a great problem would come only from the
West by individuals working at a great western university. Kumar, in other words, is
an improbability so remote as to be essentially an impossibility. Which is why, it would
seem, no number theorist would even bother reading his paper with any seriousness. Just
a different address would have ensured that he got all the help and recognition he needed
and deserved five years ago, while now he waits to be formally reviewed for a break-
through that would put his name on the front page of every newspaper in the world. He
has truly become Ellison’s invisible man.

Being Kumar’s brother I, of course, take this very personally, more so when sometime
ago he transferred all his correspondence on the RH to me, “just in case”. Even I, twelve
years younger, have started to hear the clock ticking. For him, at 74, it must be like a
drumbeat. Let us us hope that it does not come to pass, but if such a dread eventuality
were to transpire before his proof is recognised, its poignancy would be of a story from
earlier centuries, not the twenty-first.46 History would be unforgiving to the present-day
custodians of mathematics were that to happen. But the loss would not be Kumar’s
alone. The proof has direct implications for the Generalised Riemann Hypothesis, the
problem mentioned as the primary research interest of several of the Kumar’s addressees
who did not deign to respond to his emails. Who can say what dominos would then fall if
those problems are also solved? Mathematics itself awaits acknowledgement of Kumar’s
proof.

In a larger sense, this has lessons for the way scientific work should be refereed in
the 21st Century. The information age, particularly the rise of the internet, has allowed
people all over the world direct access to lectures, textbooks and research papers that
would have been impossible in earlier times except at some large and well-funded western
university. Earlier, people across the world would not even be able to learn of the refer-
ence material needed to solve a great problem, let alone have it at their finger-tips, as
now is possible. In a way, Kumar’s proof is a herald of things to come. Incredible though
it seems, the Riemann Hypothesis was the low-hanging fruit for the new possibilities
opened by the information age. With the large amount of material available on the RH
on the internet, it was just a matter of time before someone somewhere would chance
45
Kumar’s paper was sent to at least four recipients of the Bhatnagar Award in Mathematics, India’s
highest prize in Science and Engineering, with a request for reviews. None responded.
46
This sounds presumptuous as it assumes the proof is correct, but even if a fatal flaw were to be
found, that would be a normal outcome of the scientific process and not, as it now is, its unseemly arrest.

26

Page 33
upon the key lying in a darkened doorway, illuminated by a sudden flash of inspiration,
while the experts at the great universities were searching for it in the well-lit areas of the
main street...

The next great result could come from a small town in Bolivia, or a city in Nigeria.
In this new era we will discover genius is everywhere.

Afterword
The Riemann Hypothesis is notorious for its power to draw people into obsession. So it
was with me. While I understood Kumar’s work from its inception, I had no idea of the
wider problem and really had no intention of finding out. I presumed, as he did, that
he would soon enough be able to know if he was right. Or, even if wrong in the details,
he would get help, as Andrew Wiles did, to fix his proof or at least give it a form that
would constitute an advancement to the work on the RH. When that did not happen,
when Kumar quite literally hit a blank wall with his submissions to journals and cor-
respondence with experts, I, without meaning to, without even wanting to, was drawn in.

I began reading, as much as was necessary, the classic textbooks (and also Derbyshire’s
popular book [3], which helped greatly) till I understood the steps from Riemann to Lit-
tlewood, then on to Kumar’s proofs of the Random Walk, ending with the Law of the
Iterated Logarithm — and became fully convinced of the rightness of the proof. Stripped
to its essentials, it is, as I said, accessible to an undergraduate.47

It is also beautiful. The entire passage, of which Kumar’s is but a part, from the
natural numbers to the zeta function, where the peculiar and seemingly inexplicable be-
haviour of its zeros was discovered, then on to Littlewood’s great insight that converted
the inextricable problem in Complex Analysis to a simpler one of series convergence,
where again the trick was not to focus on the convergence but on the behaviour of the
Liouville integer function which was the coefficient of that series, which, most unex-
pectedly, brought us to Probability Theory, spans almost the whole of mathematics till
Riemann’s day and beyond to the probabilists of the early 20th Century, touching every
great vista on its way. I am most fortunate to have been among the first to make that
journey.

47
While the motivation to embark on this extended exercise was initially fraternal empathy, the final
conclusions were not influenced by that emotion. While tracing the entire path of the RH from its start
to Littlewood’s analysis and then onto Kumar’s steps and then finally to the LIL, which took many
months of very concentrated thinking, I was always on the lookout for a fatal flaw that may lie in the
details, and I think I came to know those details possibly better than Kumar himself did!

27

Page 34
 For this and the opportunity to contribute in my own small way, to be as T.S.Eliot
wrote,
We, content at the last
If our temporal reversion nourish
(Not too far from the yew-tree)
The life of significant soil.
I shall ever be grateful.

Vinayak Eswaran
Hyderabad, India, 21 February 2021

28

Page 35
K. Eswaran’s Comments on the Above Review of Professor Vinayak.

There is only one comment rather explanation to make. This explanation is concerned with the
Argument given by him in page 10 of his Review:

Since we are comparing the lambda sequence with a sequence of coin tosses say:
c(1), c(2),......,c(n-2), c(n-1),c(n),....
We note as follows the nth coin toss c(n) can have value +1 (H) or -1 (T) and has the properties:
(A) Obviously the value of c(n), the nth coin toss, could be +1 or -1 with equal probability
(B) The value of c(n) does not depend on the previous tosses c(n-1), c(n-2),.. etc.
In other words given any M consecutive values: c(n-M), c(n-M+1), ..,c(n-1), it is impossible to
predict the next value c(n). Properties (A) and (B) is sufficient to prove that the distance X(N)
travelled in a random walk in N steps, has the square root C0 N1/2 behavior for large N (see Eqs.
(21) and (22), Sec 5.2 of my main paper “The Final and Exhaustive Proof…”).

Since we wish to show that (n) has a similar behavior as c(n) for large n. We therefore have
proved “equal Probabilities” and also proved that (n) is not predictable if one ONLY knows the
M previous values (n-1), (n-2), ...., (n-M), for any finite (fixed) M and large arbitrary n. This
was done by proving that the relationship in Eq. (13), in his page 10, is NOT POSSIBLE. Note: In
order that we apply a strict analogy with coin tosses, it was necessary to show that without
knowing the explicit value of n, but only the M previous values of (n-1), (n-2), ...., (n-M), it
is not possible to predict the next value (n), for large n.
Hence, these results show that the  - sequence is statistically identical to a sequence of coin
tosses (or random walk) for large n and therefore the asymptotic behavior of the summatory
Liouville function L(N) for large N, will be identical to the behavior X(N), distance travelled in N
steps in a 1-d random walk. Thus proving RH. QED
We have used the fact (or rather the assumption) that mathematical logic, when used with
Peano’s Axioms in mathematical proofs, should give consistent results. So, if it can be proved
that a particular sequence {c(n)} having properties (A) and (B) will satisfy a relation R (say Eq.
(22)), then another sequence {(n)}, obeying the same properties (A) and (B) must satisfy the
same kind of relation R.

KE

Page 36
---------- Forwarded message ---------
From: Srinivasan Venkatraman <vsspster@gmail.com>
Date: Thu, Dec 10, 2020 at 2:12 PM
Subject: Re: Riemann Hypothesis paper review
To: NR RIEMANN <nrriemann@sreenidhi.edu.in>

Dear Madame,
I am hereby sending the Riemann Hypothesis paper review
Re

On Thu, Dec 10, 2020 at 7:02 AM NR RIEMANN <nrriemann@sreenidhi.edu.in> wrote:

Dear Sir,
I am writing this mail to inform you that I have not received the review of Riemann
Hypothesis paper.

Thank you sir

Regards
Dr.Suma

The Riemanns.docx
14K

Page 37
Review of Kumar Eswaran’s Papers on the Proof of RH
The Riemann’s zeta function is defined as (s)=∑⅟ⴖs. Where n is a
positive integer and s is a complex number with the series being
convergent for Re(s) ˃ 1.
The zeros of this function are at negative even intergers,-2,-4,-6…..
There are infinite number of zeros at the line at Re(s) = 1/2. The
Riemann’s hypotheses claims that these are all the non-trivial zero’s
of the function. To prove this we investigate the Liouville function
 (n) where: F(s) = ∑ (n)/ns
 (n) = (-1)͑n)֨
With Ω(n) being the total number of prime numbers of prime
numbers in the factorization of n. Then introduce the function
L(N = n=1 (n) The partial sum of (n).
The independence of (n) is shown in the sequence of(n), each
(n) is shown to be independent of the preceding ’s, for large n.
The ‫( ג‬n) is not periodic. The summatory function: L(N)= ∑𝑁 𝑛=1 ‫( ג‬n)
for the large N determines the analyticity of : F(s) = ∑ (n)/ns ,
using Littlewoods theorem and that the summatory function for the
large N mimics the random walk sequence where the sum indicates
the distance traveled from the starting point and satisfies the
relation
[c(1)+c(2) ……..c(n)] = C. N¹ʹ². The random walker behaves, for large
n, in such a manner:
1) That each step is of the same size in positive or negative direction
and each step occurs in equal probabilities +1 or -1.
2) sequence is not periodic
3) The c’s are independent of each other

Page 38
Also mod L(N) = C N¹ʹ²ᶧᵋ . Thus we see that the lambda behave like
coin toss. Thus we can use Khinchine-Kolmogrov law which enable’s
us to show L(N)/N¹ʹ²ᶧᵋ = 0 for any episilon greater than 0.
This proves the Riemann’s Hypothesis.
Judiciously using the properties of the random walk problem one
shows that Riemann’s Hypothesis is true.
There is also a numerical proof given.
I complement the author for solving the Riemann’s Hypothesis.

Professor V. Srinivasan,
Former Professor University of Hyderabad
December 10th 2020
--------

Page 39
Summary of Reviews of Papers on RH by K. Eswaran

1 Introduction
28th October 2020

Dear Committee Members,

As you know in February of this year, letters were sent by our Chairman Prof P. Narasimha
Reddy, requesting mathematicians and scientists to perform a detailed review of my work on
RH. In this report. Nearly 1200 such invitations for review along with copies of my Main paper
and links to associated papers were sent. However, only 4 persons have taken up the task of
actually doing a detailed review. Though, there was no lack of interest in my work, there have
been more than 9000 reads/downloads of my work, probably making it the most popular papers
in Number Theory in ResearchGate.

I summarize the reports of the reviewers on the above papers

There were 4 reviewers:

1) Reviewer A: Prof. Ken Roberts, Univ. of Western Ontario, Canada
2) Reviewer B: Prof. SR Valluri. Univ. of Western Ontario,Canada
3) Reviewer C: Prof. Wladislaw Narkiewicz, University of Worclaw, Poland
4) Reviewer D: Prof. German Sierra, Inst. of Th. Physics, Univ. of Madrid, Spain

The summary provided here is an outline of their comments. I have also provided more detailed
comments along with their reviews which are attached in three separate pdf files.

2. On Review by A and B
The above two reviewers jointly provided a very detailed review (in more than 20 pages) of the
papers and also wrote an Email to the Convener (Dr. A. Suma) summarizing their opinion. Both
their email and their report is enclosed along with this write up.

In their email to the convener they have said as follows:

“We found Dr. Eswaran's work quite stimulating of mathematical ideas, and believe that his work
should be brought to the attention of a wider scholarly audience, That is, the proof (or selected
portions of the methodology) should be published. The Riemann Hypothesis has resisted the
efforts of many of the best mathematicians for over 120 years,...”

My Brief comments on their review:

The reviewers have done a detailed study of the concept of towers, of equal probability and of
the λ -sequence and they have approved of the methods used. They have also they said that
there were parts of the paper that related to the details of the zeta functions which they have not

Page 40
reviewed, but since the main results utilizing these properties has been first done by Littlewood
there is no cause for worry. Another aspect they did not comment on is the use of Kechine
Kolmogorov Theorem (K-K Theorem) on the Iterative Logarithm which I used to provide the
additional (and indisputable) evidence for my proof. This was done by showing that the “width”
of the critical line must vanish to zero, hence forcing all the zeros to lie on the critical line. Here
again, the KK Theorem is too well known and has been associated with one of the greatest
mathematician of the 20th century (Kolmogorov) so there is no cause of worry.

They have also acknowledged that the extensive numerical calculations and statistical study
provides evidence of the random-walk behavior of the λ -sequence. Since the behavior of the λ -
sequence forms the heart of the proof and they have said: “we have given the first part of the
proof, the question of equal probability of the lambda sequence, thorough study”. I feel that it
can be said that the RH has been proved.

3. On Review by C
The summary of this reviewer has been extracted from an exhaustive technical discussion with
the professor. This email correspondence extended over nearly two months and dealt with the
subtitles in the proof. In the beginning the professor was skeptical and I had to explain the
procedure and intricacies of the proof in great detail and at the same time convince him of the
many objections he raised. In the end he acknowledged that I do have a proof but the method of
proof is rather unusual. Since the discussion involves nearly 60 pages of mathematics I cannot
deal with it in any detail. The entire discussion can be found in thein the attachment which
contains the entire verbatim discussion.

I only quote from the last two concluding emails (dated 14 th and 17th April), he stated that my
arguments were “heuristic”, though he says “I agree that the similarity of the considered
sequence of values of the lambda-function with a random walk gives some reasons to believe in
the truth of the conjecture”. I had replied (Apr. 16) to this saying that all I had used is
mathematical logic and the well-known properties of numbers, simply put: “It means two
mathematical entities which have similar properties will obey similar relationships” denying this,
“is equivalent to saying that mathematical logic is an unreliable tool and cannot be trusted if
implemented in mathematical proofs”.

His final reply (April 17th 2020) ended with this sentence: “I want also to stress that the word
"heuristic" has no negative meaning. A lot of work of really great mathematicians has been
performed in a heuristic way. This applies not only to old times (Euler, Laplace, the
Bernoulli’s,...) but also to recent times.”

For more details of the review and discussions the attached document may be consulted.

4 On Review by D
Unfortunately this Review is a very incomplete and in fact, if I may say so, an incorrect review.

Page 41
It all started by a misunderstanding by the Reviewer who thought that (i) I was trying to prove
the randomness of primes, a task that I did not even attempt, and (ii) that Equal Probabilities is
sufficient to prove RH, because there is no reference by him to the rest of the paper beyond the
first few pages. In fact Equal Probability is only one of the requirements for RH to be true there
are many more: the proof of unpredictability, the proof of independence and the proof that the λ -
sequence is non-cyclic. There is absolutely no reference by him to all the other theorems which
were used to prove all these requirements.

I find all this very puzzling because there was no such misunderstanding with the other three
Reviewers A, B and C. I had sent him (Reviewer D) a detailed write up not just commenting on
his review but explaining the proof and all its intricacies, (it is available as a separate attachment
along with his Review) and I suggested that he may please do a relook and a more detailed
examination, but he has not responded nor acknowledged receiving my email.

---------------

I am submitting this letter with the other documents for your kind perusal. A reading of the
comments made by the reviewers would, I hopefully believe, persuade you that the methods
and techniques adopted does provide a proof of RH and can stand on its own inspite of the
most intense scrutiny/ discussion. (E.g. the mathematical and discussion with Reviewer C, a
Polish Number Theorist covered more than 60 pages of print).

I have for your convenience written up a 10 page “Essay on My Proof”, it is available in the file
named “Comments_on_Reviewers_A_and_B_Roberts.pdf”

Regards
Dr. Kumar Eswaran
Professor/SNIST
28th October 2020

There are 4 pdf files as attachments along with this document:

1) This document: “Summary of Reviews of Papers on RH by KE.pdf”
2) My Comments and Review by A and B: “Comments_on_Reviewers_A_and_B_Roberts.pdf”
3) “Comments_on_Reviewer_C_Wladyslaw_Narkiewicz.pdf”
4) “Comments_on_Reviewer_D.pdf”

Page 42
 On Ken Roberts and SR Vallur's Review

October 27, 2020

1 CONTENTS OF THIS DOCUMENT

(1) My Comments On Ken Roberts and SR Valluri's Review Page 1

(2) Essay on My Proof of RH Page 6

(3) Email to Convenor Dr. A.Suma Page 16

(5) Prof Chandrasekar's Review of Random walk (rst 6 pages) Page 17

(6) Review by Ken Roberts and Valluri Page 23

2 My Comments on Review by Ken Roberts and

SR Valluri
I thank Professor Ken Roberts and Prof SR Valluri's for their very detailed
comments. They have taken upon the onerous task of doing a literature survey
and summarizing the work of others as well as reading my paper to great detail.

My Comments:
I will be making my comments more or less in the sequence that they had
followed.

1. Their Summary: Their summary (pages 1-2) is ne there is nothing

for me to comment about.

2. On BackGround:
This paragraph briey states the back ground of the Riemann Hypothesis.

3. On Equivalent Statements of RH:

Their writing from pages 3 to 4 is ne.
I took Littlewood's Theorem as the starting point of my proof instead of the
Equivalent statement of RH (Eq(6), p. 4). This is because when I had read
Borwein's statement attributing the equivalent statement to Landau, I could
not nd a rigorous proof. I the saw Littlewood's paper and I found that if I
choose F (s) = ζ(2s)/ζ(s) instead of his 1/ζ(s) I do get a very rigorous proof.
The Edward's book also contains Littlewood's method as given by the latter's

Page 43
published paer of 1912 (it's in French). Also, since I wanted to prove RH from
rst principles (as far as practicable), I used Littlewood's methods and applied
it to F (s) so that it is a starting point of my paper. There is no need to validate
statement 1 (their eq (6)), because it is nothing but Littlewood's Theorem
which is proved very rigorously by him by using arguments of analytical contin-
uation and has also been re-derived and thoroughly checked by me. In fact a
careful study of Littlewood's work leads us to realize that his Theorem is very
powerful, because it says that if (6) is satised then RH is True, else
it is not.
4. On Random Walks
Their write up summarizes the issues involved in the study of random walks
and its connection with RH.
In answer to the points they have raised in their 1st paragraph (of Sec 3.1)
and similar doubts in later on in this section: I only want to emphasize that
we study the statistical properties of λ(n), with our attention on Littewood's
Theorem which clearly stipulates that we need only to show that for very large
values of n (n → ∞) the λ(n) satises (i) Equal probabilities and (ii) Indepen-
dence, if this happens then the statistical distribution of the λ0 s will be identical
to the statistics of a random walk (or coin tosses). The follows from the fact
√
that to prove the | DN |=C. N behaviour of the distance DN travelled in N
steps taken randomly, it is only necessary that the random walk satises the
above two conditions (i) and (ii). (I am attaching the rst 3 pages of the Nobel
Laureate Prof Chandrasekar's famous Review paper which contains only these
√
two conditions (i) and (ii) in the derivation of | DN |=C. N behaviour, I have
annotated the pages 3 and 4 for the convenience of the reader.)
In another question the reviewers asked if there is any other condition that
√
the λ0 s need satisfy for the N relationship to hold? The answer to it is that
only (i) and (ii) need be satised. If there is some other condition property it
is not necessary for the proof.
To justify my statement I have to take recourse to the inherent assumption
of the consistency of mathematical logic in mathematical proofs. I argue as
follows: If a particular sequence {a(n)}satises two properties viz Property P
and Property Q and if by using the axioms of arithmetic and mathematical Logic
we can legitimately deduce that {a(n)} satises Property R, then any other
sequence say,{b(n)} which satises Property P and Property Q must satisfy
Property R. You will agree with me that any other result would call to question
the legitimacy of Mathematical Logic in matematical Proofs.

5. Borwein Integrals and and Random Walks

The write up about Borwein Integrals was interesting reading.
Much of it was new to me. I am familiar with the use of Sinc function
Sin(x)/x in the solution of diraction problems. But I am not very sure that
these Borwein Integrals can directly help in analysing the discrete λ-sequence,
because λ(n) is an arithmetic function and hence the arithmetic properties of
numbers have to be used to discover the behavior of the sequence.

6. Alternative Expression for F(s)

Page 44
 I was aware of the statement that the terms of a series which is not absolutely
convergent can be rearranged to get what ever result one wants. But I did not
know, till now, that it was Riemann who proved it. Because of the dangers
pointed out by this Rearrangement Theorem, I was very careful in deriving
a new representation of F (s) making sure that I start from its dened series
representation by choosing choose some s such that Re(s) >> 1 ensuring that
it was not convergent but absoulutely convergent.
They have also stated that: We have worked through the details of that
example. We are comfortable with this rearrangement of the Dirichlet series
representation of F(s) via the Liouville function, because it is being done when
everything is absolutely convergent, that is for Re(s) > 1. And they noted that
the justication of methods used to obtain the representation can be found in
Titchmarsh's book theory of Functions.

7. Def of Towers P (m; p; u) and Alternative Expression for F (s)

They have very nicely explained the concept of Towers and its use in the
alternate expression of F (s).
They have taken the trouble to go into great detail explaining each step. I
would like to state that by doing so they have travelled almost the exact mental
route that rst led me to realize that a proof of RH is possible! After writing
Eq.3.10) (in my paper) I immediately realised that in each sub-series ('Towers')
the λ0 s alternate in sign and therefore the +10 sand −1s have to be equal. Then
by using this new expression of F (s) to analytically continue the function to
the regions to the left of the vertical Re(s) = 1 line, in the complex s-plane,
by using Littlewood's methods, I obtained an equivalent expression for L(N ) as
depicted by the Fig 1. This gure was crucial to my understanding
1 because a
careful examination convinced me that it is the random behavior of the λ(n)0 s
for large n which explains that all the poles of F (s) and hence the nontrivial
zeros of ζ(s) lie on the critical line Re(s) = 1/2.
8. A Brief Intro On the various proofs of Equal Probability.
Actually, it so turns out that the Prime Number Theorem proves Equal
Probability. This proof of course is not mine. I have given two other proofs one
is the method of Towers and the other by using the method of constructingn-
integers using multiplication of primes and then using induction. So there are
actually 3 proofs for Equal Probability. I will write about each of them in turn.

In order to satisfy Littlewoods criteria we need to prove Equal Probabilities

of the only for very large n (near or at innity ).
λ(n)0 s
λ(n) can take only two values viz +.1 or -1. And the sample is the
Since the
+
whole set of positive integers Z . And since we are interested only in N tending

1 As an after thought, I feel all the things about towers and even the Figure could have be
avoided. One can greatly reduce the length of the proof (and the paper) if one just sticks to
(1) the def of F(s), and its use in (2) Littlewoods Theorem and (3) prove equal probability
and (4) prove Independence to show its statistical similarity to a random walk (coin tosses)
and nally (5) invoke the Khinchine- Kolmogorov iterative logarith to lanny prove RH. This
procedure would have given a crisp and sharp proof of RH (much like what Gauss would have
perhaps done) but would have camouaged all the insights which led to the proof.

Page 45
to innity (or at innity).
2 There is no need to use complicated denitions of
probability and probability spaces. Our denition of probability is simply to
answer the question: if we choose an arbitrary integer n (where n can be any
number upto innity),what are the chances that λ(n) will be +1 or equal to -1.
The equal probabilities theorem can be proved in many ways. (a) By using
3
the Prime Number Theorem , (b) by the method of Towers and (c) My the
method of Induction,
In the main paper is by forming Towers (every number will be in only one
unique tower) and then showing that for each number n with λ(n) = +1, there
is an unique number m (it's twin) with λ(m) = +1.This would imply that Equal
Probabilities is true. If not, suppose that P rob(λ = 1) = 0.75, this means that
0
there are an insuciency of integers with their λ s= −1. This is impossible
because every number has twin number whose λ-value is it's opposite.
So we no go back to your comments.

8(i) On Equal Probability via PNT (your para 4.1, p 12)

It has been shown that the Prime Number Theorem when interpreted in
λ(1)+λ(2)+λ(3)+...+λ(n)
terms of λ0 s means:
n → 0 as (n → ∞) (See Borwein etal
4
in footnote ). Now since we also know that zeta function has no zero on the
line Re(s) = 1. L(N ) < c.N a+ ,
We also know from Littlewood's theorem that
now, since we know there is no zero at Re(s) = 1 we are permitted to put a = 1
L(N )
and  = 0, which leads to
N = c, (N → ∞), since c is nite (else there will
a pole of F (s) at Re(s) = 1, meaning a zero of ζ(s) at Re(s) = 1). Therefore,
Equal Probability Theorem follows.
The above should clear your doubts that you expressed in paras 3 and 4.
The answer is that the PNT does prove Equal Probabilities.
We now move to the proof of Equal Probabilities by Towers.

8(ii) Equiprobability via Towers Argument (para 4.3 p 21)

The concept of towers has been explained in the Main Paper and its ap-
pearence in the alternative expression for F (s). The Equal probability follows
from the fact that each integer is in a unique tower whose members are in an

2 It will be very useful to give a meaning to the word at innity. What we do is: to the
number system of positive integers Z + ≡ {1, 2, 3,4,.......} we add the number ∞and call this
augmented set Z∞. (Just like in geometry, to the ordinary 2-D Eucledian space A2 we add
the point at innity to get the projective space P^2.) By this denition conceptualization
becomes much simpler. (To the purist we can arbitrarily dene λ(∞) = −1.)
3 Frankly, at the time of rst writing in September 2016, K.Eswaran: The Dirichlet Se-
ries for the Liouville Function and the Riemann Hypothesis. I was not familiar with the
fact that the PrimeNumber Theorem (PNT), can lead to the proof of Equal Probability.
I had seen a paragraph in Borwein's book (see Ref[3] in my Main Paper; also see slide 32
in Peter Borwein's Lecture on RH ) stating that Landau had proved that the statement
λ(1)+λ(2)+λ(3)+...+λ(n)
n
→ 0 as (n → ∞) is equivalent to the PNT and this leads to the
Equal Probabilities. However, at that time (2016) I could not nd a rigorous proof of Equal
Probabilities and that is why I took it upon myself to prove it from rst principles by the
methods of Towers.
4 Peter Borwein, Stephen Choi, Brendan Rooney, and Andrea Weirathmueller, 2008, The
Riemann Hypothesis: A Resource for the Acionado and Virtuso alike, Canadian Mathe-
matical Society, 2008

Page 46
ever increasing sequence of integers with alternating λ-values and hence equal
number of +10 s and −10 s.
In their comments the Reviewrs A and B gave an accurate description of the
building of Towers and its deployment to prove Equal Probability.

8(iii) Equiprobability (My 2nd Proof)

I wanted to give a 3rd proof of Equal Probability in such a manner that one
need not take recourse to mappings or pairings, but only use induction. This
is because many purists consider mathematical induction (especially when used
in countable sets) as a more fundamental proof. ( I agree that this is a matter
of opinion).

I only use the intutive fact that for every odd number there is a unique even
number (its successor) and that in the sequence of natural numbers, odd and
even alternate therfore the number of odd numbers is equal to the number of
even numbers.
We start our demonstration by building up the set of integers
{Z ∞ } = {1, 2, 3, 4, 5, .....} not by addition (as is normally done), but by
multiplication of powers of primes. Consider
5 {2} = {1, 21 , 22 , 23 , 24 , ...}.We
0
see that the λs of each integer in the set alternate i.e equal nos. of +1's
and -1's . Hence, 3.{2} = {3.1, 3.21 , 3.22 , 3.23 , 3.24 , ...} again equal nos. of
2 2 2 1 2 2 2 3 2 4
+1's and -1', we can again go on 3 .{2} = {3 .1, 3 .2 , 3 .2 , 3 .2 , 3 2 , ...},
which again has equal nos. of +1's and -1'. By continuing and then dening
{3} = {1, 3, 32 , 33 , 34 , ...}. We get all the set of integers factorizable by 2 or 3 or
0
both viz. {3}.{2} = {2}.{3}will have equal nos. of +1's and -1' in their λ s . The
same situation holds for: {2}.{3}.{5}, which consists of all integers factorizable
by 2 or 3 or 5 or by any combination of their products. We can obtain all integers
by writing {Z ∞ } = {2}.{3}.{5}.{7}.{11}...which will similarly have equal nos
0
of +1's and -1's in their λ s. This proof employs induction: eg. if {2}.{3}.{5}
0
has equal +1's and -1's in their λ s, then {2}.{3}.{5}.{7} has equal +1's and
0
-1's in their λ s. And Equal Probabilities must follow (anything else leads to a
contradiction). QED
It may be asked: Why do I need so many proofs? My answer is that since
this is a proof of RH, my earnest desire is to put all doubts and objections
beyond the pale.

9. The Rest of the proof of RH

I have written down a brief essay on RH below which will, I am sure help in
explaining the rest of the proof. This has been done for the benet of the Lay
person who wishes to have an overview of the proof.

The following Note (next page) is a stand alone 'Essay ' on the crucial
theorems and steps that lead to the proof, I have included it so that any person
could help you to undersatnd the rest of the proof.
xxxx-
5 Notice that in the exponents of {2}the number of even and odd exponents are equal and
also that their λ- values alternate as +1 and -1.

Page 47
3 Essay on my Proof of RH
You may have noticed that I make no attempt to prove the randomness of
primes (a conjecture which is very dicult to prove and, luckily for me, is not
required for the proof of RH). What I use in my proof is, the randomness of
the sequence of λ(n) (which appears as the nth term in the Liouville series),
in the sense that the λ-sequence (series) resembles a random walk, in that (a)
λ(n) = −1 or +1 are equally probable and (b) λ(n) is not dependent on previous
values of theλ's upto any nite distance, as the length N of the sequence tends
6
to ∞. This statistical resemblence to a random walk for large N, persists even
though some individual members of the λ's are related by the deterministic
relation λ(mn) = λ(m).λ(n).
These considerations, I then show is enough to prove RH, using the Khinchin
and Kolmogorov's Law of the Iterated Logarithm. Both (a) and (b) can be
proved in more than one way and I have done so in my paper. Actually, as
I have shown, the fact that λ's are deterministic and obey the deterministic
relation λ(m.n) = λ(m).λ(n), and therefore not random in the classical sense,
does not matter
7 at all for large N. If (a) and (b) can be proved then RH is
proved.
8
There are three things which are very important to the proof. (1) Littlewoods
theorem (2) Equal probabilties and (3) Independence. Since (1) and (2) have
been already dealt with in the preceding, I will not dwell much on them except
to say that I had given two proofs and that at the time of writing that PNT
could also be used to povide a rigorous proof of Equal probabilities. It was only
later, I became aware that if one uses the Prime Number Theorem and the fact
that ζ(s) has no zero on the vertical line Re(s) = 1 and also Littlewoods theorem
(stated in my paper) it is possible to prove Equal Probabilities see slide 238...in
9
Ref [6] (below)Vinayak's lecture . I have also justied the PNT proof in para
8(i) preceding. Since we seem to have accepted that (a) Equal Probabilities is
proved, what remains to prove is (b) that, the λ(n) is independent of previous
λ0 s (within a nite distance) in a long sequence of length N → ∞. Therefore
they would closely resemble the statistical behavior of a ramdom walk (coin

6A careful study of Prof Chandrasekar's paper, will make one realize, that to determine
the characteristics of a random walk all that Chandrasekar used are two assumptions (i)
Equal probabilities and (ii) Independence in a sequence of coin tosses. He requires no other
assumption, Hence, since (i) and (ii) are the same as (a) and (b), if these can be proved for
large n λ- sequence, the required result will follow.
, this time for the
7 For λ(n) for some n, the formula λ(m.n) = λ(m).λ(n),can determine the
example, given
next predictable value λ(2.n) = λ(2).λ(n) = −λ(n), but since we consider very long sequences
(length N → ∞) for a large n, say n = 10100 , the integer 2n will be at a distance of 10100 from
n making such a prediction statistically insignicant! This situation is even more so for the
integers in a tower. For example, if there is a large integer j which is in the tower P (m; p; u)
and can be written as say j = m.pk .u then the very next integer (in the tower) above j is the
integer m.pk+1 .u which is even more far away because generally p >> 2.
8 There are a lot of redundancies in my rst paper which, I now think, could have been
omitted altogether.
9 Vinayak Eswaran (my brother), had taken the trouble to write out 7 lectures on my proof
of RH, see Ref [6]. The lectures are well worth reading because of their clarity, lack of jargon
and assumes only the pre-reqisite of an under-graduate level of mathematical foundation.

Page 48
toss).
The independence of λ(n), for large arguments n,i.e. condition (b), is shown
in two dierent ways. In the rst (see Sec 11.3 Appendix IV of Ref.[1]) while it
is acknowledged that the λ0 s are linked through the multiplicative relationship
λ(mn) = λ(m).λ(n) etc., it is shown that the numbers linked in this manner
move increasingly further from each other so that this distance tends to innity
as n → ∞. Thus in the sequence {λ(n)} each λ(n) is independent of preceding
λ0 s which lie within a nite distance from it as n → ∞.
In the second proof of Independence (see para 2(a), p 21, in Ref[1]), it is
shown that any functional relationship which binds λ(n) to its previous λ0 s
(which lie within a distance L, arbitrary but xed) and which is valid for all
n,would make the sequence λ(n) periodic after some large n > n0 . It is then
proved that the sequence λ(n) cannot have such a cyclic behavior, because this
would imply, from Littlewood's theorem, that there are no zeros of the zeta
function within the critical strip - which is not true.
With the properties (a) and (b) proved, a direct application of the Khinchine-
Kolmogorov Law of the Iterated Logarithm shows that the RH is TRUE and
that the width of the critical line is zero!

In the next section onwards I outline the main steps of the proof, which will
be a useful to read the paper. (I also strongly recommend Vinayak's Lectures
Ref [6] in the Reference Section (below), the lectures provide a detailed, and
sometimes an alternative, version of my proof ).
In the last I have added an experimental Verication which is not a part of
the proof. But the extensive calculations verify the properties of the lambda's
over very large sequences and hence lend support and credence to our methods.

4 Gist of The Proof (See Ref. 3 for Details)

The proof proceeds in essentially 4 basic steps.
STEP 1: We choose
10 an analytic function, F (s) = ζ(2s)/ζ(2s), whose
poles exactly correspond to the non-trivial zeros of the zeta function ζ(s). F (s)
is analytic in the region Re(s) > 1and is given by:

∞
X λ(n)
F (s) = (1)
n=1
ns

STEP 2: An analytical continuation of F (s) to the left of the vertical Re(s) = 1,

using Littlewood's Theorem determines that the asymptotic behavior of the

10 It may be mentioned here that for his study Littlewood had chosen the function 1/ζ(s)
which had lead to the µ-sequence. A diculty with using this Mobius function is that µ(n)
can take values−1, 0, or + 1, whereas λ(n) takes values of −1 or + 1, like a coin toss. Because
of this it is easier to compare the λ-sequence with coin tosses rather than the µ-sequence. And
because of this very reason I chose to study the function F (s) = ζ(2s)/ζ(s) , which leads to
the λ-sequence.

Page 49
summatory function L(N ) :
N
X
L(N ) = λ(n) (2)
n=1

as N → ∞ plays a crucial role in determining the analyticity of F (s) and the

position of the poles of F(s) and thereby the zeros of ζ(s) in the critical region
0< Re(s) <1. Littlewood's theorem states that the asymptotic behaviour of
L(N ) for large N, determines the analyticity of F(s), and if the behaviour is
such that

N
X
| L(N ) | ≡ | λ(n) |< C N a+ (f or large N ) (3)
n=1

(where (1/2 ≤ a < 1 ), and  is a small positive number), F(s) will be analytic
in the region (a < Re(s). This is a very crucial result as far as RH is concerned
because if one can determine that actually a = 1/2 in (3) then the Riemann
Hypothesis is proved.
STEP 3: In this step we logically put forth the argument: that the very
necessity that (3) must be satised for the Riemann Hypothesis to be true, im-
poses very severe resrtictions on the behaviour of the sequence of the Liouville
functions: {λ(1), λ(2), λ(3), .....}. These restrictions (conditions) are given later
in this section.
The λ(n) λ(1) = 1and for n > 1 : λ(n) = (−1)Ω(n) and
is dened as:
is determined by factorizing n and nding Ω(n), the number of prime factors
of n (multiplicities included). We already know λ(n) is fully determined by
factorizing n and is an arithmetic function namely: λ(m.n) = λ(m).λ(n), for all
integers m, n.
Now for RH to be true one must have a = 1/2 in Eq.(3), the rst N terms
(N large) of the λ sequence must therefore sum up as:

| λ(1) + λ(2) + λ(3) + ...... + λ(N ) | ' C . N 1/2 (4)

The above equation brings to mind a similar relationship satised by another

sequence c(n) = ±1, which corresponds to the nth step of a One-dimensional
random walk! (This c(n) can be simulated by coin tosses, if we replace Heads
by +1 and Tails by -1; so a N-step random walk can be thought as a coin toss
experiment where a coin is tossed N times.) It is well known that for such
a random-walker's sequence the sum indicates the distance travelled from the
starting position in N steps and satises the relationship:

| c(1) + c(2) + c(3) + ...... + c(N ) | ' C . N 1/2 (5)

The well known result of Equation (5), (see S.Chandrasekar), is derived by

using the assumption that the random walker behaves in such a manner that:
(i) Each step is of the same size but can be either in the positive direction or
negative direction i.e the nth step c(n) can be +1 or -1, with Equal Probability.

Page 50
 (ii) The sequence of steps cannot be periodic, that is the pattern of steps
cannot form a repetitive pattern (there is no cycle)
11 .
th th
(iii) Knowing the n step the (n+1) cannot be predicted. That is, knowing
c(n), c(n+1) cannot be determined (they are independent).

The above assumptions are enough to derive Eq(6). This has been shown
by many researchers (e.g. See S. Chandrasekar, refered in Ref[1])

4.1 The Argument:

Comparing (4) and (5) leads us to deduce some inevitable conclusions:

Eq.(4) must be satised by the λ(n) sequence if the Riemann Hypothesis
is TRUE, this is the conclusion that we deduce from Littlewoods Theorem,
with the proviso that Eq. (4) needs be satised only for large N (this being
the condition of Littlewood's theorem). Now, there are many Random walks
possible, for instance: 100 random walkers can each of them, take N steps and
each of these random walkers will be at anapproximate distance of distance
C . N 1/2 from the starting point. Each of these 100 sequences can be thought of
as 100 dierent instances of a random walk of N steps each.
If we wish to compare (5) with (4) there are several conceptual issues: (α)
The sequence in (4) is a deterministic sequence and (β) we have only one se-
quence. We get over this latter issue by considering the single sequence as one
instance of a hypothetical random walk of N steps. And even though the λ(n)'s
are deterministic (an aspect we temporarily ignore) we could investigate this
one instance and argue (or hypothesize) that when N is large, the following
rules could be obeyed:
Properties of the λ−sequence
(a) Given an arbitrary large n chosen at random, there is Equal probability
of λ(n)being either +1 or-1.
(b) The λ−sequence cannot be periodic, that is the λ(n) cannot form a
repetitive pattern (no cycle)
(c) Knowing the value of λ(n) it is not possible to predict λ(n + 1) . Unpre-
dictability (independence).

Note the rules (a),(b)and (c) are similar to (i),(ii) and (iii) and therefore:
If by using the number theoretical (arithmetical) properties of the integers, the
primes and the factorization process, it is somehow possible to prove that the
λ(n) satisy the rules (a), (b) and (c) then just as (5) is satised by every instance
of a random walk, Eq(4) will be satised for our one particular instance of our
λ−sequence and thus RH will be proved! Taking this as a cue we proceed. NOTE:
According to Littlewood's Theorem: It is only necessary that the λ − sequence
satisy the rules for very large lengths of the sequence and large arguments of λ.
It is because of this relaxation provided by Littlewoods theorem that even though
the λ−sequence is deterministic, but its behaviour still very closely approximates
11 The requirement that there are no cycles was not necessary for Chandrasekar, but it is
necessary for the proof of (iii) i.e. independence.

Page 51
to the statistical behaviour of a sequence of random walks (or coin tosses) over
large N.
Hence the next step is to prove the properties for large values of N i.e. when
N tends to innity. (It will also become clear later that the deterministic nature
of the λ(n)0 s, does not signicantly disturb the above statistical properties.
12 )
However, we are actually now at the crossroads: we have to prove that
the λ−sequence possesses the above properties (a),(b) and (c) for large sequence
lengths and large arguments N . Property (a) has already been proven, it is quite
possible that by using the artithmetic properties of the λ(n) that (b) can be
proved (as has been done in the Appendix III of the Main Paper) but it is in
the proving of (c) that the real diculty lies. This is because all mathematical
proofs in Arithmetic relies heavily on the Axioms of Peano (P.A), but P.A.
does not come to our aid for certain hard problems e.g to prove (or disprove)
that the advent of primes are random. Luckily we don't have to decide upon
this last surmise! But, we do have to decide upon the problem of resolving
what is independence or unpredictability of a sequence. So we dene that
a sequence {a} ≡ {a(1), a(2), .....a(n), ....} as unpredictable, for large values of
its arguments, if when given the value of a(N ) where N is large and the M
previous values where M << N (and M nite), viz.{a(N − M ), a(N − M +
1), ....., a(N − 1), a(N )} then it is not possible to predict a(N + 1). It can be
easily seen that if a sequence {a} has this property then since, it is not possible
to predict the value of a(N +1) knowing a(N ) and its M previous values, we can
assert that a(N ) and a(N + 1) are independent. It will be proved that by this
denition the λ−sequence ≡ {λ(1), λ(2), ....., λ(N ), ....}the components of λ(N )
and λ(N + 1), are independent
13 for large values of N. Using this knowledge we
can proceed.

STEP 4: Proof of the Properties of the λ−sequence. In this step several

theorems are proved using the number theoretical (arithmetical) properties of
integers, primes and the unique factorization of integers to establish the prop-
erties (a), (b) and (c) of the λ−sequence as listed in the previous paragraphs.
These proofs are fairly straight forward and are from rst principles:
We have already seen that Property (a) On Equal Probabilities, is proved

12 See Foot note 1, on page 1.

13 I dene that a sequence {λ(n)} is not Independent : If there is some n0 and some M (both
nite) such that for every k > n0 one can predict λ(k) given its M previous values (Note n0
and M can be very large but must be nite. Also the requirements of Littlewood's theorem
are such that these properties need hold only for large n (or k) tending to ∞). That is, there
exists some function f (x1 , x2 , x3 , ..., xM ) such that one can write λ(k) = f (λ(k − 1), λ(k −
2), λ(k − 3), ....λ(k − M )). If such a function indeed exists, then one can use it to obtain λ(k)
and recursively calculate λ(k + 1), λ(k + 2),..., for all higher values of n in λ(n), this will make
the {λ(n)} predictable. I follow Kurt Godel (See K.Godel: (1931) On formally undecidable
propositions of Principia mathematica and completeness and consistency, pp 592-617, See
p. 601 to 602, In Jean van Heijenoort's book From Frege to Godel, Harvard Univ Press
(1967)), by using the concept of recursive functions in our denition of independence. Kurt
Godel had said that anything that can be computed (in our case predicted) can be represented
by recursive functions. Once the reasonableness of these denitions are accepted then it is not
dicult to prove, by using the fact that integers are factorizable uniquely and the arithmetic
properties of the λ(n) that the λ−sequence satises Equal Probability and Independence.

10

Page 52
in Theorems 2 and 3 in Section 5.2, in the Main Paper Ref [1] and that the
concept of towers is used in the proofs. An alternative proof
14 by constuction
of all prime products and induction is also given in Ref [2]. A third proof, which
follows from Littlewood's theorem but assumes the fact that there is no zero
with Re(s)=1, (proved in the Prime Number Theorem) can also be derived as
has been discussed in the foregoing and also by you (but is not given in the
paper).
Property (b) On no cycles, is proved in Appendix III, Ref [1]
Property(c) On unpredictability (independence) is proved in Appendix IV,
Ref[1]. An alternative proof of this also given: See para 5(a), in page 2, of Ref
[3].
An alternative arithmetical proof of the asymptotic relation | L(N ) |' c =
C.N 1/2 . is given in Appendix V, Ref [1].
15

We have therefore showed that Eq. (3) is satised by the λ−sequence. We

will now get an expression for the width of the critical line and show that this
width vanishes in the limit of large N.
We have established the fact that the λ−sequence behaves like a coin tosses
(or a random walk) and this entitles us to use Khinchine -Kolmogorov's law
of the Iterated Logarithm, adapted to the present context, is: Let {λn } be
independent, identically distributed random variables with means zero and unit
variances. Let SN = λ1 + λ2 + ... + λN . The limit superior (upper bound) of
SN almost surely (a.s.) satises
SN
Lim Sup √
2N loglogN
= 1 as N → ∞
Now, from Theorem 4 we have written that if we consider the λ0 s as coin
1
+dN
tosses one can write | L(N ) |=| λ(1) + λ(2) + ... + λ(N ) |≤ C0 N 2 (as
N → ∞) (since we are interested in only the behaviour for large N we henceforth
ignore the constants). Comparing this expression with the one above we see
1 √
that one can write N 2 +dN ∼ N loglogN thus yielding an expression for dN =
logloglogN
. We see that dN → 0 as N → ∞ , this satises the equivalent
2logN

14 This alternative proof of equal probabilities is given in Ref [2] and is done by explict
construction of integers by products of sets which are powers of a given prime. Each of these
sets are seen to be comprised of members of ascending magnitudes and alternating λ−values,
in perfect analogy with the ascending and alternating odd/even sequences of the natural
numbers. Thus the member n of each set has the exact probability of 1/2 (as n → ∞) of its
λ−value being +1 or −1 as the natural numbers have of being odd or even. As every natural
number can be placed uniquely in one such subset, it is seen that a randomly chosen natural
number will have a probabilty of 1/2 of having its λ−value equal to +1 or −1(as n → ∞).
15 In a separate arithmetical study Ref [4], it was discovered that for very large N, smaller
primes contribute more (than the larger primes) to the calculation of λ(n)'s which occur in the
summatory expression for L(N ). Specically, if one chooses an integer K such that K << N
then the primes p which are s.t p < N/K , occur much more often in the calculation of each
term in L(N ) than the large primes q which are s.t. N/K < q < N . This situation permits
us to deduce, interestingly, that if we allow both K and N → ∞ in such a manner that the
CK
ratio N/K is a xed number, then we must have: P r(λ(n) = 1 | n < N ) = 1/2 − and
logN
CK
P r(λ(n) = −1 | n < N ) = 1/2 + logN where CK is a small uctuating number which tends
to zero as K → ∞; thus once again conrming that the L(N ) behaves like a random walk for
very large N.

11

Page 53
 L(N )
statement of (3) viz. 1 = 0, for any chosen >0 (Ref. Eq.(20) in Ref.[1])
N 2 +
. Further d∞ ≡ (LimN →∞ dN ) is the half-width of the critical line. Since this
is zero, we conclude that all the non- trivial zeros of the zeta function must lie
strictly on the critical line.
Thereby, the Riemann Hypothesis is proved.

5 Experimental verication
In the last Appendix VI, of the Main Paper, Ref[1], numerical experiments
(using Mathematica) are described and there it is shown that large sequence of
lambdas behave like a random walk (or equivalently like coin tosses).
I have calculated consecutive λ(n)0 s forming a large sequence, denoted by
Λ[N0 , M ], of length M of the form Λ[N0 , M ] ≡ {λ(N0 ), λ(N0 + 1), λ(N0 +
2), ..., λ(N0 + M − 1)}.where N0 is some large integer.
It has been shown by actually computation and performing a χ2 t using the
methods suggested by Donald Knuth,
16 that these very large sequences contain-
ing consecutive values of λ(n)0 s very closely resemble and are in fact statistically
indistinguishable from a Binomial distribution (coin tosses) of equal length.
I have done very many computations (using Mathematica) and some of them
have been presented in the Tables in Appendix VI,sec.3, e.g. Tables 1.3 and
1.4 page 27; also see End Note on the last page of this document. These are
accurate and actual computations and the numerical results are indisputable
17 .
By very many numerical computations I have shown that the sets of consec-
p √ p
utive λ0 s denoted as S+ (N ) = Λ(N +1, N ) and S− (N ) = Λ(N − N +1, N )
, (N a square integer) have the property of being statistically indistinguishable
from coin tosses.
18
These sequences called S− (N ) and S+ (N ) exist (N being a perfect square)
and behave like random sequences (coin tosses) and the concatination of such
sets of S− (N ) and S+ (N ) cover all of λ(n) for all integers n upto innity. This
shows that the entire {λ(n)} sequence is made up of an innite series of subse-
quences of type S− (N ) and S+ (N ) each of which statistically behave like coin
tosses! The Tables 1.4 cited above, provide ample proof of this. The purpose of
this section is just to demonstrate that the, predictions of the theorems proved in
the Main Paper, have been numerically veried extensively. The verications
19

16 Knuth D.,(1968) `Art of Computer Programming', vol 2, Chap 3. Addison Wesley

17 I believe my papers provides the raison d'etre for the existence of this phenomena.
18 The reason for this was demonstrably argued because each integer n occurring in the
argument of λ(n) S+ (N ) belongs to a dierent Tower.
in one of the sets say
Notice that if you choose N = j 2 then the union of the two sets :
S+ (j 2 + 1, j) ∪ S− ((j + 1)2 − j, j + 1) is nothing but the sequence
{λ(j 2 +1), λ(j 2 +2), λ(j 2 +3), ..., λ((j +1)2 )}. That is they cover all the λ0 s with arguments
between two consecutive perfect squares, j 2 to (j + 1)2 . Now if you choose N = (j + 1)2 you
can cover the next region between the perfect squares ((j + 1)2 + 1) to (j + 2)2 and therefore
you can capture all the regions between two consecutive perfect squares by concatinating such
sets all the way up to innity - basically covering all integers by the union of sets S− and
S+ right up to innity.
19 In fact I have proved (in Appendix VI of the Main Paper) that if you do a χ2 t of

12

Page 54
have been done by doing a χ2 t of a λ−sequence with a Binomial distribution
(coin tosses). In every case it has been shown that for large N the λ−sequence
is indistinguishable from a random walk (sequence of coin tosses).
These numerical computations and χ2 correlations are very real and are ac-
tually present and give very strong indications of randomness present in the
λ(n)0 s which were actually computed by the factorization of integers n. I
strongly believe this phenomena has to be explained by the Pure Mathemati-
cians and not brushed aside or put under the carpet or carelessly labeled as mere
coincidence!
I wish to emphasize, that I have not only explained this phenomena but also
showed how it connects with the proof of the Riemann Hypothesis.

6 Conclusion
In this write up I have shown that the reason for the RH to be true lies with
the fact that the λ- sequence behaves statistically like coin tosses. It was shown
that a sequence c(n) of coin tosses or a sequence of a random walk, exhibits the
square root behavoir of Eq(4), was deduced by Khinchine-Kolmogorov, Chan-
drasekar and others, from the assumption of two criteria (i) Equal Probabilities
(ii) Independence. By using the properties of arithmetic and from the use of
mathematical deductions we could prove many theorems (and many have alter-
native proofs) to show that the λ- sequence for large values of its arguments
also satiisfy (i) and (ii). And therfore they satisfy Littlewood 's condition for
RH to be true. We also made extensive numerical computations involving the
λ−sequence; all these support the various theorems we have proved. Since,
we have started from rst principles and used only the arithmetic properties
of numbers and mathematical logic to prove the theorems, in my opinion, this
leaves hardly any doubt as to the truth of RH. I believe that I have solved it
comprehensively.

K. Eswaran

7 References
[1] The nal and Exhaustive proof of the Remann Hypothesis...

[2]A Simple Proof That Even and Odd Numbers of Prime Factors Occur
with Equal Probabilities in the Factor-ization of Integers

[3]A Quick Reading Guide to the Proof of the Riemann Hypothesis

[4] The eect of of the non-random-walk behavior of the Liouville Series

L(N) by the rst nite number of terms.

a sequence of P
λ0 s of length N with
√ a sequence of coin tosses of equal length (using Knuth's
method) then N n=1 λ(m+n) = χ N for N large. If necessary, one may consult the extensive
numerical calculations done at the end of the last Appendix involving λ−sequences as long as
100,000 and argumets of λ as large as 10, 000,000,000!

13

Page 55
 I enclose below the slides of the Invited Lecture that I delivered at the
Government Arts & Science College Kumbakonam on March 1st 2019. (This
was followed by another (slightly shorter) Lecture delivered in the Ramnujan
Centre of Sastra University on the evening of the same day).
[5] Invited Lecture On the Riemann Hypothesis by K.Eswaran

[6] Vinayak Eswaran: Seven Lectures of Kumar Eswaran's Proof on RH

K. Eswaran/Professor
Sree Nidhi Institute of Science and Technology,
Yamnampet, Ghatkesar, Hyderabad 501301
10th October 2020

8 END NOTES -These are just end notes which

-provide some more information.
Just for curiosity I tested the behaviour of very long sequences of lambda and
compared them with coin tosses by χ2 tting.See the Appendix VI of my Main
Paper Ref[1], there are many more examples in the form of Tables.
PK+L √
Here we dene the summation: SUM ≡ n=K λ(n) = χ L (see Eq(9), in
page 25 of Ref[1]).
Everywhere the χ2 ts get better and better as the the length of the sequence
and the size of the integer increases. Knuth speculated that a value of around
4 or 5 to make the sequence indistinguishable from a sequence of coin tosses or
a random-walk. However, Littlewood's criterion is far less strict, it is sucient
PK+N √
that as N tends to innity: n=K λ(n) = C N , where C can be any nite
value.
In the tests below the lambda sequence passes the test in every case. I
have taken very Long sequences, for Example III, I have considered 100,000
consecutive integers starting from K = 25 × 1024 + 1
EXAMPLE I: A sequence of Length L of consecutive lambdas starting from
λ(25000001) to λ(25005000) of 5000 λ values for 5000 consecutive integers,starting
from 25, 000, 001 ie L = 5000. We use Mathematica commands in our compu-
tations as shown below.
Plus[LiouvilleLambda[Range[25000001, 25005000]]]
= {1, 1, 1, 1, 1, 1, 1, -1, -1, -1, 1, -1, -1, 1, 1, 1, -1, -1, 1, 1, \ 1, -1, -1, 1, 1,
-1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, 1, -1, 1, -1, \ -1, -1, 1, -1, -1, -1, -1, 1, -1, -1, 1,
-1, 1, -1, 1, -1, 1, 1, 1, \TRUNCATED-- \ 1, -1, 1, 1, 1, 1, 1, 1, -1}
SUM= -42, χ2 = [SUM * SUM/ L] = [42*42/5000}= 0.3528

EXAMPLE II
Starting Integer K = 25 × 1024 + 1; L = 10, 000
Plus[LiouvilleLambda[ Range[25000000000000000000000001, 25000000000000000000010000]]]
SUM= 88, χ2 = 88*88/10000 = 0.7744

EXAMPLE III:
Starting Integer K = 25 × 1024 + 1; L = 100, 000

14

Page 56
Plus[LiouvilleLambda[ Range[25000000000000000000000001, 25000000000000000000100000]]]
SUM= 238, χ2 = 238*238/ 100000 = 0.56644

EXAMPLE IV:
Starting Integer Z = 1030 + 1; L=1000;
Plus[LiouvilleLambda[Range[Z, Z + 999]]]
SUM= -20, χ2 = 20*20/1000 =0.4

EXAMPLE VI:
Starting Integer Z = 1030 + 1; L=10,000;
Plus[LiouvilleLambda[Range[Z, Z + 9999]]]
SUM= 54 χ2 = 54*54/10000 = 0.2916

15

Page 57
9 Email to Convenor Dr. A. Suma from Ken
Roberts
Ken Roberts <krobe8@gmail.com> to: NR RIEMANN <nrriemann@sreenidhi.edu.in>
cc: Sree Ram Valluri <vallurisr@gmail.com>, "Dr. Kumar Eswaran" <ku-
mar.e@gmail.com>
Oct 21, 2020, 9:15 AM

Dear Dr. A. Suma,

We are pleased to forward a report, with our remarks on Dr. K. Eswaran's
proposed proof of the Riemann Hypothesis. Our report is document PPRH-
20201021 dated 21-Oct-2020. This report is joint work of Prof. S. R. Valluri
and myself.
Our report is incomplete, in that we did not examine all aspects of the
proposed proof. There are some aspects of the Riemann zeta function with
which we are not suciently familiar in order to speak authoritatively, However,
we have given the rst part of the proof, the question of equal probability of the
lambda sequence, thorough study.
We found Dr. Eswaran's work quite stimulating of mathematical ideas, and
believe that his work should be brought to the attention of a wider scholarly
audience, That is, the proof (or selected portions of the methdology) should
be published. The Riemann Hypothesis has resisted the eorts of many of the
best mathematicians for over 120 years, and any advance, even a partial one,
is to be communicated. We believe that Dr. Eswaran's towers construction,
for instance, is quite innovative and useful. As well, we believe that his paper
may prompt the exploration by other scholars of some related topics, including
pseudo-random-walk sequences and the Borwein integrals seen as random walks.
Details of those suggestions are in the attached report. We are not suggesting
that Dr. Eswaran must personally investigate those topics, as they are not
directly required for the renement and validation of his proof. Rather, such
related topics represent an opportunity for anciliary investigations by the wider
mathematical community.
We appreciate the opportunity to review and comment upon Dr. Eswaran's
proposed proof. We hope that all aspects of the proposed proof are correct
or can be amended to resolve uncertainties. Even if the proposed proof turns
out to be decient in some manner, it does constitute an useful advance on the
Riemann Hypothesis problem. As well, though we have not addressed the topic
much in our attached report, we found some of Dr. Eswaran's representations
of the behaviour of RH-related sequences to be quite insightful and thought-
provoking. Those studies also deserve to be shared with a wider audience.
Best wishes, and thank you, Ken Roberts 21-Oct-2020
copies to: Prof. S. R. Valluri and Dr. K. Eswaran attachment: PPRH-
20201021 (pdf le) ATTACHMENT CONTAINING THE DETAILED REVIEW
IS IN A SEPARATE PDF FILE

16

Page 58
 BY PROF. S CHANDRASKAR

For your convenience, I have copy pasted only the first three Pages CHAPTER ONE of Prof
Chandrasekhar’s Paper:

CONTINUED NEXT PAGE

Page 59
.

Page 60
Page 61
For your Convenience I have cut and pasted, below, Prof Chandrasekhar’s Appendix I.

Page 62
CONTINUED ON NEXT PAGE

Page 63
End of Appendix.

Page 64
 Remarks on a Proposed Proof
of the Riemann Hypothesis
=================================

Ken Roberts1 and S. R. Valluri2

October 21, 20203

Summary

This document has our remarks on Dr. Kumar Eswaran’s proposed

proof of the Riemann Hypothesis. These remarks address the proof’s
mathematical ideas and methods. We focus on the aspects of KE’s
proposed proof which are innovative. We do not attempt to place the
proof in the context of historical and recent other work on the RH.
We accept well known results and adopt the notation and terminology
which is used in KE’s writings. We find it advisable to elaborate upon
some aspects for clarity. We pose a number of questions. Some ques-
tions highlight a statement which we have not been able to adequately
verify from the cited literature. These are really requests for further
detailed information such as location of the statement within a cited
document. Other questions call attention to a potential shortcoming
in some reasoning. These are requests for further clarity, and are not
necessarily critical to the main argument of the proposed proof. Fi-
nally, some questions call attention to related topics. These questions
are suggestions of possible opportunities for further interesting explor-
ations by KE or others.

KE’s writings which were consulted are [KE1, KE2, KE3]. The primary
paper is [KE1] and the other two papers provide some background. A
convenient brief overview is [KE2] which makes reference to [KE1] for
details and justifications. Another helpful reference on the proposed
1
Email: krobe8@uwo.ca or krobe8@gmail.com, Physics and Astronomy, Western University,
London, Canada.
2
Email: vallurisr@gmail.com or valluri@uwo.ca, Physics and Astronomy, Western University,
and Kings University College, London, Canada.
3
File location: LT4:u4:kwork4:eswaran

PPRH-20201021 Page 1 21-Oct-2020

Page 65
proof is a series of seven slide lectures by Dr. Vinayak Eswaran, which
are available either at Youtube or as a pdf document [VE1].

KE’s proposed proof is quite innovative, and has several interesting

ideas. These methods may also be useful for other problems. We are
at present undecided on whether the proof is accurate in all respects.
There are some aspects which we believe require clarification in order
to construct a fully justified proof. We have found the study of KE’s
proposed proof quite stimulating. It suggests additional topics for
investigation, and we appreciate the opportunity to consider KE’s work
in detail.

At present, we have made a detailed examination of only a part of the

proposed proof, the portion which deals with the question of whether
the λ(n) values (defined below) have equal probability of being +1 or
-1. We are forwarding this (incomplete) review at this time, in the
hope that it will be useful for Dr. Eswaran and also for others who are
examining the proof.

In our opinion, the proposed proof should be published so that it may

be examined by a wider community of scholars. Even if the proposed
proof turns out to be deficient in some manner, it does constitute an
advance on the RH problem, which has been an open question and
investigated by numerous experts for over 120 years. Any advance
in methodologies is worthwhile as a contribution to the investigation
of the Riemann Hypothesis. We believe in particular that Dr. K. Es-
waran’s towers method is insightful and suggests further opportunities.

In the remainder of this document, we will summarize the proposed

proof, and highlight questions which we have. We do not necessarily
expect Dr. Eswaran to take responsibility for addressing all questions.
We hope that, with wider publication, other scholars will be drawn to
this interesting subject as there are many opportunities for work on
the details.

PPRH-20201021 Page 2 21-Oct-2020

Page 66
1 Background

Peter Borwein and colleagues have prepared an excellent 2008 hand-

book on the Riemann Hypothesis [PB] which contains a summary of
prior work and a selection from the many important papers related to
the RH. Another important resource is the 1974 book by H. M. Ed-
wards on the Riemann Zeta Function [HME]. Each book also includes
an English translation of B. Riemann’s original research report of 1859
in which he first stated his hypothesis.

Riemann considered a function ζ(s) of a complex variable s which, if

Re(s) > 1, is given by the Dirichlet series
∞
1 1 X 1
ζ(s) = 1 + s + s + ... = s
. (1)
2 3 n=1
n

Riemann extended the ζ(s) function by analytic continuation to the

whole complex plane, except for the point s = 1. The ζ(s) function has
zeros when s is an even negative integer. Those are called the trivial
zeros. As well, ζ(s) has an infinite number of isolated zeros s = σ + it.
All known non-trivial zeros of ζ(s) lie on the line σ = Re(s) = 12 . That
line is called the critical line. The Riemann Hypothesis (RH) is that
all the non-trivial zeros of ζ(s) lie on the critical line.

Two of the early serious investigations of the RH were by T. J. Stieltjes

in the 1880s and by J. Hadamard in the 1890s (see [HME], pp. 262-
263). Subsequent investigators include many of the great names in
number theory and analysis.

2 Equivalent Statements to RH

KE’s paper [KE1], following one of the avenues taken in prior invest-
igations, defines a function F (s) by
ζ(2s)
F (s) = . (2)
ζ(s)

PPRH-20201021 Page 3 21-Oct-2020

Page 67
That function has poles at exactly the non-trivial zeros of ζ(s) and is
convenient to study.

Chapter 5 of the handbook [PB] has several statements which are equi-
valent to the RH. First we need some definitions.

Define a function Ω(n) on the natural numbers, letting Ω(n) equal the
number of prime factors dividing n, counting multiplicity. For example,
Ω(40) = Ω(23 · 5) = 3 + 1 = 4.

Define the Liouville function λ(n) via the formula

λ(n) = λn = (−1)Ω(n) . (3)

The Liouville function (LF) is completely multiplicative. The notations
λ(n) and λn will be used interchangeably, as convenient.

The relation between the function F (s) defined above and the Liouville
function is that F (s) has the Dirichlet series with terms given by the
LF: ∞
X λ(n)
F (s) = s
. (4)
n=1
n
Thus one can approach the Riemann ζ(s) function via a study of the
properties of the Liouville λ(n) sequence, and vice versa.

Define the summatory Liouville function L(N ) via the formula (see
[KE1], equation (1.2b)),
X
L(N ) = λ(1) + λ(2) + ... + λ(N ) = λ(n). (5)
n≤N

The notations L(N ) and LN will also be used interchangeably.

It is known (see [PB], Equiv 5.2, pg. 46) that the RH is equivalent to
Statement 1: For any fixed  > 0,
 L(N ) 
lim 1 = 0 as N → ∞. (6)
N 2 +

PPRH-20201021 Page 4 21-Oct-2020

Page 68
3 Random Walks

KE’s insight is that the Liouville function λ(n), can be considered

as an unbiased random walk with steps ±1. That is not a new ob-
servation. There is considerable numerical evidence that λ(n) looks
random. However, KE has given the logic of the idea the serious and
detailed consideration which it deserves. The mathematical challenge,
as outlined in [KE2], is to develop arguments which make it clear that
the properties observed in the expected behaviour of unbiased ran-
dom walks as a class, are also present in the particular deterministic
sequence λ(n), and that those properties are sufficient for the λ(n)
sequence to satisfy the analogue of Statement 1 above.

Let c(n) denote an infinite sequence of ±1 values. Let S(N ) denote

the corresponding summatory function,
X
S(N ) = c(1) + c(2) + ... + c(N ) = c(n). (7)
n≤N

Suppose that the sequence c(n) is an unbiased one-dimensional random

walk with steps ±1. In random walk terminology, the summatory
function S(N ) is the position of the particular walker c after having
taken N steps. That c(n) is unbiased and random means that, for any
particular n index, the probability of c(n) being +1 is 1/2. Moreover,
the value of c(n) does not depend upon the values of c(k) for the n − 1
values k = 1, 2, ..., n − 1. With these assumptions, the probability
distribution function of S(N ) is a binomial distribution with a mean
of zero. The second moment |S(N )|2 of √ that distribution is N , so
the dispersion, the RMS value of S(N ), is N . That means the c(n)
sequence satisfies
Statement 2: For any fixed  > 0,
 S(N ) 
lim 1 = 0 as N → ∞. (8)
N 2 +

PPRH-20201021 Page 5 21-Oct-2020

Page 69
If one can show that λ(n), despite being deterministic, behaves “suffi-
ciently like” an unbiased random walk c(n), then it will satisfy State-
ment 2, and hence Statement 1 will be established, and thence the
Riemann Hypothesis will be established. That is the plan of the proof.

3.1 Pseudo-Random Walks and Deterministic Sequences

It is worthwhile to give more thought to the concept of how a de-

terministic sequence might be “sufficiently like” a random walk. At
various places in the text of [KE1] and related papers, the assertion
is made that any sequence c(n) (deterministic or not) which shares
certain properties with unbiased random walks will necessarily satisfy
Statement 2. We are not confident of that assertion. It requires
further justification, in our opinion. It would be a useful result, and
deserves to be explored fully. It presumably derives from a careful
working through of the proofs which are customarily given for random
walks, with the deductions from randomness of the sequence replaced
by appropriate precise conditions on the sequence.

We make the following remarks and ask a variety of questions, in order

to stimulate exploration of the topic:
– The essence of the transition from random walks to the properrties
of a deterministic sequence, likely consists of moving away from the
ideas of “randomness” and “probability” towards “statistics”. That
is, it is desirable to be able to show that the λ(n) and L(N ) values
possess certain statistical properties, and that any sequence with those
statistical properties will have the property required for Statement 2
to be true of it.
– Can there be sequences c(n) which do not satisfy Statement 2?
Certainly. But they may not be unbiased (will have nonzero mean)
or may contain periodicities or correlations between finite-scope sub-
sequences (hence not be random). Or ... ?? In what other ways might
a sequence fail to satisfy Statement 2?
– Suppose that we identify certain properties which is it reasonable to
expect of a random walk – such as those mentioned above. Imagine

PPRH-20201021 Page 6 21-Oct-2020

Page 70
that such a sequence is deterministic, produced by some algorithm.
Call such a sequence a “pseudo-random-walk” (of some type, depend-
ing upon the properties which have been chosen to designate that
type of pseudo-RW). Can one show that ALL pseudo-RW sequences
of that type will satisfy Statement 2? For instance, is there perhaps
a summability-type argument, and an associated Tauberian theorem,
which would let one conclude that property of a specific sequence from
the behaviour of an ensemble of sequences?
– The properties of a pseudo-random sequence which are asserted to
cause it to satisfy Statement 2 are, for instance (see [KE1], section
5.2, point 2 in the discussion of the λ(n) sequence) that the values of
the c(n) are ±1 with equal probability, and that the values are non-
cyclic. Are there other properties required? For instance, the absence
of bounded-length cycles may not be enough to ensure that for each
n value, the value of c(n) does not depend upon the values of c(k) for
the n − 1 values k = 1, 2, ..., n − 1.
– It should be understood that, in these remarks, we are not neces-
sarily asking whether the specific sequence c(n) = λ(n) might satisfy
Statement 2. Rather, we ask whether any non-cyclic equiprobable se-
quence of ±1 must necessarily satisfy Statement 2. Is it possible that
some particular additional feature of the λ(n) sequence is also essen-
tial, such as being completely multiplicative, in order for any non-cyclic
equiprobable c(n) to satisfy Statement 2?
– It might be, in fact, interesting to explore sequences c(n) which are
completely multiplicative but satisfy Statement 2. There should be
a family of such sequences, for example by setting c(p) = 0 for some
subset of the primes and c(p) = ±1 for other primes.
– A general study of pseudo-RW sequences may be worthwhile. Like
most mathematical topics, someone has probably already investigated
this, though perhaps not with the perspective brought to the task via
these RH investigations. So one asks, what work has already been
done on pseudo-RW sequences?
– For a proof of the RH, this desired result (that ANY pseudo-RW se-
quence satisfies Statement 2) is not strictly necessary. The sequence
λ(n) has a particular structure, being completely multiplicative, that
may suffice. The details of lecture 6 of [VE1] may suggest some ideas
of properties of a sequence c(n) that might be relevant. We are more
confident of reasoning which is also based upon multiplicative prop-

PPRH-20201021 Page 7 21-Oct-2020

Page 71
erties of the sequence, rather than an assertion regarding all possibly
deterministic sequences with equal probabilities and independent sub-
sequences.
– Attempting to construct counterexamples may be an clarifying exer-
cise. That is, to take some pseudo-RW sequence (of a certain type, ie
satisfying well defined conditions) and, since we are allowing it to be
deterministic, modifying it in such a way that it retains the properties
characterizing its type, but has other properties rather different from
a pure random walk.

That is rather a scattershot list of questions, with some overlap and re-
petition. We pose these questions partly to indicate some opportunities
for interesting explorations which might come from a consideration of
the details in KE’s proof. However, that list is also an indication of
some discomfort on our part. Without a better grasp of at least some
of those topics, we are not fully convinced that the reasoning in the
proposed proof is complete.

3.2 Borwein Integrals and Random Walks

We also wish to mention another topic, relevant to the proof structure,

prompted by the towers construction illustrated in Figure 1 of [KE1].

The Borwein integrals have an interpretation in terms of random walks.

See [DJB] for the original paper which defines the integrals which came
to be called the Borwein integrals. See the various references listed at
[WBI] for background on these integrals, including the relationship of
these integrals to random walks.

The Borwein integrals are Fourier cosine transforms of a finite product

of scaled sinc functions. The function sinc(x) equals sin(x)/x if x
is nonzero, and sinc(0) = 1 for continuity. For a > 0, define the
modified characteristic function χa of the interval [−a, a] by χa (x) = 1

PPRH-20201021 Page 8 21-Oct-2020

Page 72
if −a < x < a, χa (±a) = 1/2, and χa (x) =p 0 for x outside [−a, a]. The
Fourier cosine transform (FCT) of χa is a 2/π sinc(ax). Conversely,
the FCT of the latter function is equivalent to χa . We can also interpret
χa as the probability distribution of a random variable which makes
one step, of length uniformly chosen between 0 and a > 0, equiprobably
in either a positive or negative direction.

Suppose that a random walker takes two steps, each step being equi-
probably in the plus or minus direction, and of length uniformly between
0 and a, and between 0 and b. The ending point is the convolution of
the functions χa and χb . The FCT of that probability distribution for
a two-step walk is an integral of the product of two sinc functions.

If the random walker takes n steps of lengths uniformly chosen between

0 and ak , equiprobably in plus or minus directions, for k = 1, ...n, then
the probability distribution of the walker’s ending point is an integral
of the product of the n sinc functions.

The essential fact about the Borwein integrals is that, if the sum of
the maximum step lengths is sufficiently short then the FCT integral,
that is the integral of the product of the sinc functions, equals π/2.
However, if the sum of the step lengths is long enough, then the integral
of the product of the sinc functions is less than π/2, by a rational
number factor C which can be exactly calculated.

We do not want to take up too much space in these remarks, but give
the following example, which led to the paper [DJB]. The paper has
much more than just this example, and suggests many opportunities
for further exploration.

Here is the example:

For n = 0, 1, ..., define an = 1/(2n + 1), and sn = a0 + a1 + ... + an .
Define τn by
Z ∞Y n 
τn = sinc(ak x) dx. (9)
0 k=0

PPRH-20201021 Page 9 21-Oct-2020

Page 73
Then τn = π/2 for n = 0, 1, 2, 3, 4, 5, 6 but τn < π/2 for all larger n.
That is, written out explicitly,
Z ∞  π
τ0 = sinc(x) dx = , (10)
0 2
Z ∞  π
τ1 = sinc(x) sinc(x/3) dx = , (11)
0 2
Z ∞  π
τ2 = sinc(x) sinc(x/3) sinc(x/5) dx = , (12)
0 2
... and so on up to n = 6 ...
Z ∞  π
τ6 = sinc(x) ...sinc(x/13) dx = . (13)
0 2
However, for n > 6, we have τn < π/2, as seen for the case n = 7,
Z ∞  π π
τ7 = sinc(x) ...sinc(x/13) sinc(x/15) dx = C7 < . (14)
0 2 2
The factor C7 in τ7 is given by
(w − 1)7
C7 = 1 − 6 , (15)
2 · 7! · w
where
1 1 1
w= + + ... + . (16)
3 5 15

The interpretation of that property in terms of random walks, is this:

Because 1/3 + 1/5 + ... + 1/13 is less than 1, a random walk of 6 steps
of lengths up to those maxima will lie within the interval [−1, 1]. How-
ever, appending an additional step of length up to 1/15 will produce
a total walk length which may lie outside the interval [−1, 1].

The result is quite general. By making the steps shorter, one can
obtain more sinc integrals in the product before the pattern breaks
down.

Consider the diagram in figure 1 of [KE1]. It shows an interpretation

of the summatory Liouville function L(N ) in terms of a random walk
with equiprobable steps of length ±1.

PPRH-20201021 Page 10 21-Oct-2020

Page 74
Remarks and Questions:
– There is perhaps a relationship between the concept of the summat-
ory Liouville function L(N ) as the ending point of a random walk, and
the interpretation of Borwein integrals as random walks.
– It is unlikely that the Borwein integrals can be used to produce a
counterexample to the Riemann hypothesis. The RH has been ex-
plored to such an extent, via zero finding, that a simple example such
as a Borwein integral with steps say of 1/p where the p values are
primes, is probably not the source of a difficulty. Nonetheless, what
might be the structure of an attempt to use the Borwein integral effect
to construct an RH counterexample?
– Otherwise posed, the question becomes: What is it about the Bor-
wein integrals represntation of the probability distribution of the end-
point of a random walk, and the Liouville function or other represent-
ation of a random walk which allows a random walk based upon, say,
steps of size 1/p to succeed, ie comply with the Riemann Hypothesis?
Is it finiteness of the integrand product in the Borwein integrals? What
about infinite analogues of the Borwein integrals?
– What happens if we imagine transforming some Liouville function
representation via an analogue of the Fourier cosine transform?
– There are related papers, linked via the Borwein integral references,
mentioned above, that bear on such questions.
– Whittaker and Watson [WW], section 6.24, exercise 6 on page 122
discuss these integrals . The result is partly due to Störmer.

We emphasize that our interest in the Borwein integrals was stimulated

by KE’s proposed proof of the RH, and in particular his intriguing fig-
ure 1 in [KE1]. It represents an opportunity for related investigations.
We are not expecting those investigations to be done as part of the
validation of the proposed proof of the RH.

4 Equal Probability of λ(n) Values

The first objective in KE’s proof is to establish that the λ(n) values
have an equal probability of being +1 or -1. There is strong numerical

PPRH-20201021 Page 11 21-Oct-2020

Page 75
evidence for that claim, but what is required is a math-logic validation.
Ideally the proof will be stated in statistical terms, not probabilistic
terms, in order to avoid ambiguity related to invoking the concept of
probability for a deterministic sequence.

The proof of equal probability which is most detailed and also, in

our opinion, the most satisfactory of the alternative proofs presented,
is the towers proof. That proof uses a partitioning of the natural
numbers into subsequences called “towers” (see [KE1], first couple of
pages of section 5.2) in a clever rearrangement of the natural numbers
to ensure alternation of the λ(n) values because successive n values
within the rearrangement have alternatively an odd or even number of
prime factors.

In a moment we will move on to the argument for equal probability

which uses towers. However, we first wish to mention a theorem on
rearrangements of conditionally convergent series, which as it happens
is due to Riemann.

4.1 Riemann’s Theorem on Rearrangment of Series

Riemann’s theorem is described in section 28 of Bromwich’s book on

Infinite Series, page 74 of the second edition or page 68 of the first
edition [BIS], and in section 44 of Knopp’s book on the Theory of
Infinite Series [KTIS], page 318.

Riemann’s Series Rearrangement Theorem: Suppose a series

of real numbers converges to a finite sum, but it does not converge
absolutely. Choose any real value S. Then the terms of the series can
be rearranged, so that it converges to the sum S. It is also possible
to rearrange the terms of the series so that it diverges to positive
or negative infinity, or so that it oscillates between any two finite or
infinite limits.

PPRH-20201021 Page 12 21-Oct-2020

Page 76
One proof involves dividing the terms of the series into two ordered
sets, the negative terms in one set and the positive terms in the other
set, with the zero terms placed in either set. Because the series is con-
vergent, the terms in each set tend towards zero. Because the series
is not absolutely convergent, the terms in each set add to respectively
negative and positive infinity. The result follows via choosing a re-
arrangement of the terms which makes the partial sums oscillate on
either side of the desired limit, or close to desired lim-inf and lim-sup
values. The resulting rearrangment does not alter the relative order
of the negative terms, or the relative order of the positive terms, but
merely changes the manner in which the two ordered subsets are in-
terwoven.

Remarks and Questions:

– Series which are convergent but not absolutely convergent are called
conditionally convergent.
– This result indicates that a single conditionally convergent series
can be rearranged to achieve any desired behaviour. However, it is not
clear that a series whose terms depend upon a parameter, either a con-
tinuous or perhaps a discrete parameter, can be rearranged to achieve
an approximation to a target function or values. That is closer to the
situation with the λ(n) series since multiplicity imposes constraints.
Alternatively, if considering the F (s) function via its Dirichlet series,
one has a parameter s in play which imposes other constraints.
– The availability of the parameter s will be important in the equi-
probability proof via towers of [KE1], which will be considered shortly.
The proof via towers works because there is a function F (s) whose rep-
resentation via towers involves a rearrangement of the Dirichlet series.
That rearrangement will preserve the sum, for Re(s) > 1, because the
Dirichlet series of F (s) is absolutely convergent for such s. Hence with
a bit of extra reasoning (done in [KE1]) one gets a representation of
F (s) with alternating signs of the λ(n) numerators. Details to be dis-
cussed below.
– Modifying a conditionally convergent series by adding an absolutely
convergent series, or omitting a subseries of terms which are absolutely
convergent, will not alter the fact that the series is conditionally con-
vergent.

PPRH-20201021 Page 13 21-Oct-2020

Page 77
4.2 Definition of Towers P (m; p; u)

Now we turn to the towers proof of equal probabilty. For convenience

we are writing P (m; p; u) for towers instead of the notation Pm;p;u of
[KE1]. Also, we will write (m, p, u, e) for the label which is assigned to
a natural number n ≥ 2, rather than the notation (m, pe , u) of [KE1].
Section 2 of [KE1] defines towers, and Appendixes 1 and 2 provide
additional justification of the definitions.

Natural numbers with one or more prime factors of multiplicity greater

than 1 will be considered class I. In contrast, numbers with no prime
factors of multiplicity 2 or more will be considered class II. Class II
numbers are called square-free numbers elsewhere in the literature. We
do not know a short word for numbers which are not square-free, so
will simply call those non-square-free. Given a number n, we will call
the primes which appear in the factorization of n with multiplicity 2
or more the multiprimes of n, and the primes which appear in the
factorization of n with multiplicity 1, the uniprimes of n. Class I is
the non-square-free numbers, which have one or more multiprimes.
Class II is the square-free numbers, which have no multiprimes, only
uniprimes.

If n ≥ 2 is non-square-free (class I), let p denote the largest multiprime

of n, and let e denote the multiplicity of p in n. The value of p is used to
split the prime factors of n into two subsets, lesser primes and greater
primes. Let m denote the product of all lesser prime factors of n (to
their appropriate multiplicities if they are multiprimes in n), and let
u denote the product of all greater prime factors of n. All the prime
factors in u are necessarily uniprimes, since p is the largest multiprime
of n. Then the class I number n is represented by n = mpe u. That is
equation (2.7) in [KE1]. The label (m, p, u, e) is attached to n.

If n is square-free (class II), then define p as the largest prime factor of

n, define m as the product of all lesser prime factors of n, set u = 1 and
set e = 1. All the factors of such an n are uniprimes, so the product

PPRH-20201021 Page 14 21-Oct-2020

Page 78
m is also square-free. The label (m, p, 1, 1) is attached to n. The same
equation applies, n = mpe u for a class II number, and in this case
u = 1 and e = 1 so n = mp.

Every n ≥ 2 has thus been assigned a label (m, p, u, e).

The tower P (m; p; u) is defined to be the ordered set of all numbers n

with labels (m, p, u, e) for any value of e. Otherwise visualized, each
natural number n ≥ 2 lies in a 4-dimensional space along m, p, u, e axes,
and the tower sets P (m; p; u) are sets of points in the 3-dimensional
(m, p, u) space obtained by projecting (m, p, u, e) along the e-axis. The
elements of the tower are considered to be in ascending order.

Some towers are empty sets. For instance, if m has a prime factor
bigger than or equal to p, or if u has a prime factor lesser than or
equal to p, then there is no tower P (m; p; u). Or rather, that tower is
an empty set.

Furthermore, if u > 1, then any n in P (m; p; u) must be a multiprime,

because a square-free n would have been placed in a tower whose label
has u = 1. Also, if m has a multiprime factor, then all numbers in the
tower P (m; p; u) must be non-square-free (class I), because they are
multiples of m.

The tower P (m; p; u) will be given by either equation (2.8) or equation

(2.9) of [KE1], which are respectively

P (m; p; u) = {mp2 u, mp3 u, mp4 u, ...} and (17)

P (m; p; 1) = {mp, mp2 , mp3 , ...}. (18)

Are these two equations consistent? KE verifies the consistency, and

we agree. Any inconsistency would arise if u = 1, and concern the
tower into which a number n = mp would be placed. That would only
happen if m is a product of primes less than p, and each such prime

PPRH-20201021 Page 15 21-Oct-2020

Page 79
occurs in m with exponent 1. In such a situation, the square-free value
n = mp is class II, and appears as the first element of the tower
P (m; p; 1) = {mp, mp2 , mp3 , ...}. (19)

Supposing m a product of uniprimes less than p, the other equation

for u = 1,
P (m; p; 1) = {mp2 , mp3 , mp4 , ...}, (20)
would be incorrect. That tower P (m; p; 1) also includes the class II
number n = mp, so is given by
P (m; p; 1) = {mp, mp2 , mp3 , ...}. (21)

Thus, the towers are well defined. The towers which contain a class
II number contain only one such number, at the start of their ordered
set. Such towers are P (1; p; 1) which is all the powers of a prime p, and
P (m, p, 1) which contain a product of m (formed from primes smaller
than p, all primes in m being uniprimes), times all the powers of p.
Those are the towers which show up in equation (3.10) of [KE1] with
their summation index starting at r = 1 instead of k = 2.

With these definitions every number n ≥ 2 is in exactly one tower, and

all towers are disjoint. Each tower’s sequence of increasing numbers
has alternating values of λ(n), starting with either +1 or -1 depending
upon the particular values of λ(m) and λ(u) and the initial power of
p which appears in the tower: that is, λ(p) = −1 or λ(p2 ) = +1.

4.3 Alternative Expression for F (s)

The function F (s) = ζ(2s)/ζ(s) has its poles at the non-trivial zeros
of the Riemann zeta function, and has as its Dirichlet series
∞
X λ(n)
F (s) = . (22)
n=1
ns

PPRH-20201021 Page 16 21-Oct-2020

Page 80
That series, the Dirichlet series of the Liouville function, is absolutely
convergent for Re(s) > 1. Section 3 of [KE1] derives an alternative
expression for F (s). The expression involves a rearrangement of terms
in the Dirichlet series. The rearrangement is valid if Re(s) > 1 since
the Dirichlet series is absolutely convergent for such s.

Using towers, the paper [KE1] presents a representation of the Dirichlet

series with the summation rearranged into a sum of subseries. Each
subseries is a summation over a tower. Each tower has alternating
values of λ(n), so the probability of λ(n) being say +1 in a particular
tower is exactly 1/2.

The towers construction and the rearrangement of the Dirichlet series

to an alternative expression are best understood via the example which
is shown in equation (3.10) of [KE1]. We have worked through the
details of that example. We are comfortable with this rearrangement
of the Dirichlet series representation of F (s) via the Liouville function,
because it is being done when everything is absolutely convergent, that
is for Re(s) > 1. Dirichlet series are unique (more precisely stated in
Theorem P2 below), so that any Dirichlet series which also represents
F (s) for Re(s) > 1 must be the same λ(n) series. That is the meaning
of the statement towards the end of section 3 of [KE1], that the h(n)
numerators in the Dirichlet series of equation (3.16) equal the λ(n)
values, as per equations (3.18a,b,c).

Remark: The uniqueness of a Dirichlet series representation of a func-

tion can be found in Titchmarsh’s Theory of Functions (see [TTF],
section 9.6, pg 309) or in Hardy and Wright’s Theory of Numbers (see
[HW], section 17.1, page 245).

Theorem P2: If ∞ ∞
X an X bn
= , (23)
n=1
ns n=1
ns
for s in some region (open set) of values of s, then an = bn for all values
of n.

PPRH-20201021 Page 17 21-Oct-2020

Page 81
4.4 Representation of Summatory Liouville Function L(N )

Most of section 4 of [KE1] is concerned with the explanation of figure

1 of the paper. We found that very interesting and suggestive. The
figure is not part of the proof but it provides some clarity, and suggests
ideas for further exploration. Our remarks made earlier, regarding the
Borwein integrals interpreted as random walks, were partly prompted
by figure 1. KE has a similar insight, as illustrated by his mention
of figure 1 being a representation of L(N ) by rectangular waves. Ex-
ploration of this topic is likely worthwhile in general, not just for its
immediate relevance regarding a proof of the Riemann hypothesis.

The last paragraph of section 4 mentions section 6 of the paper [KE1],

which is presumably a reference to section 5.2.

The rearrangement of order of the terms in the Dirichlet series for F (s)
is a bit unusual. Each tower is an infinite sequence, so the representa-
tion is a sequence of sequences. All terms being absolutely convergent
for Re(s) > 1 is the condition which allows such an unusual rearrange-
ment.

A question which arises, is whether the equal probability result for

the λ(n) sequence is correct in a statistical sense, on finite segments
[1, N ] of the natural numbers. For each N , define N pos to be the
count of n values satisfying 1 ≤ n ≤ N fpr which λ(n) = +1. Define
Q(N ) = N pos /N to be the fraction of n values in [1, N ] for which
λ(n) = +1. The statement that P rob{λ(n) = +1} = 1/2, means
that, as N → ∞, Q(N ) → 1/2. A rearrangement of the sequence,
as such, does not necessarily imply that Q(N ) tends to a limit. It
is conceivable that Q(N ) might oscillate between two distinct values
(lim-inf and lim-sup) instead of tending to a limit. Either a rigourous
proof or an attempt to construct a counterexample seems in order.

If “equal probability” is not defined as in the prior paragraph, then

we might be working with a new concept. Traditional random walk

PPRH-20201021 Page 18 21-Oct-2020

Page 82
reasoning might not be applicable. The towers rearrangement is inter-
esting, and suggests a way of moving forward towards a proof of the
RH. However, it may be necessary to consider a new variety of random
walk, one which is carried out on an infinite ordered set.

This is also a concern because the rearrangement of the series F (s)

represents F (s) as a sum of zero-hugging subsequences (the towers).
Each tower is a deterministic walk which never gets more than ±1 step
away from zero.

The solution to this is no doubt to utilize special properties of the λ(n)

sequence, for instance its relation to the primes, and also that λ is a
multiplicative function.

5 The Remainder of the Proof

We have not examined the remainder of the proposed proof in detail.

The most important point we have not considered is whether the λ(n)
values are non-cyclic. As we understand the point, the question is
whether there can be a finite-length relationship which allows one to
obtain the value of λ(n) from knowledge of a bounded number of pre-
decessors – for instance, from λ(n − k), λ(n − k + 1), ...λ(n − 1), where
k is a fixed number and n is allowed to become arbitrarily large. We
wonder, however, that even if that point is answered in the negative
(ie, the λ(n) sequence is non-cyclic), whether there can still be other
relationships among the λ values which might prevent it from being a
random walk in terms of statistical properties.

We have not given the remainder of the proof in [KE1] the detailed
consideration that it deserves. We have, however, identified some ques-
tions which are worth further investigation. For instance, regarding
the Borwein integrals, some queries which were inspired by the paper
[KE1].

PPRH-20201021 Page 19 21-Oct-2020

Page 83
We hope that all aspects of the proposed proof are correct, or can be
amended to resolve uncertainties. The paper has inspired us to some
new opportunities. For that reason we believe the methods (towers,
eg) of the [KE1] paper should be published so they are made known
to a wider community of scholars. We are advancing our remarks and
questions in this report with the sincere hope that they will stimulate
other constructive discussion and effort towards proving the Riemann
Hypothesis and exploring various related topics such as the Borwein
integrals and pseudo-random-walks.

We appreciate the opportunity to review and comment upon Dr. Es-

waran’s proposed proof.

References

[DJB] David Borwein and Jonathan M. Borwein, 2001, “Some Re-

markable Properties of Sinc and Related Integrals”, The Ramanu-
jan Journal, vol 5, pp 73-89, 2001.

[PB] Peter Borwein, Stephen Choi, Brendan Rooney, and Andrea

Weirathmueller, 2008, The Riemann Hypothesis: A Resource for
the Afficionado and Virtuoso Alike, Canadian Mathematical So-
ciety, 2008.

[BIS] T. J. I’a. Bromwich, 1908, An Introduction to the Theory of

Infinite Series, First edition, Macmillan & Company, 1908. Second
edition, Macmillan & Company, 1926, reprinted 1965.

[HME] H. M. Edwards, 1974, Riemann’s Zeta Function, Academic

Press, 1974. Reprinted by Dover Publications, 2001.

[KE1] K. Eswaran, May-2018, “The Final and Exhaustive Proof of the

Riemann Hypothesis from First Principles”,
www.researchgate.net/publication/325035649

PPRH-20201021 Page 20 21-Oct-2020

Page 84
[KE2] K. Eswaran, March-2019, “The Pathway to the Riemann Hy-
pothesis”,
www.researchgate.net/publication/331889126
[KE3] K. Eswaran, April-2018, “A Simple Proof that Even and Odd
Numbers of Prime Factors Occur with Equal Probability in the
Factorization of Integers and its Implications for the Riemann
Hypothesis”,
www.researchgate.net/publication/324828748
[VE1] V. Eswaran, October-2019, “Seven Lectures on Kumar Es-
waran’s Proposed Proof of the Riemann Hypothesis”. There are
seven video lectures at
www.youtube.com/channel/UCuLORNpknTzLB7s57-htzcw/
videos
Each video lecture displays a sequence of slides, without audio.
The slides as a pdf file are available at
www.researchgate.net/publication/336899740
[HW] G. H. Hardy and E. M. Wright, 1938, An Introduction to the
Theory of Numbers, 4th edition, Oxford University Press, 1960.
[KTIS] Konrad Knopp, 1921, Theory and Application of Infinite
Series, 4th German edition, 1947. 2nd English edition, Blackie
& Sons, 1951. Reprinted by Dover Publications, 1960. Note that
[KTIS] is not the same as Knopp’s Infinite Sequences and Series,
which was also published by Dover. Knopp’s “Theory” book is
three times as long and has an abundance of advanced informa-
tion.
[TTF] E. C. Titchmarsh, 1932, The Theory of Functions, 2nd edition,
Oxford University Press, 1939.
[WW] E. T. Whittaker and G. N. Watson, 1927, A Course of Modern
Analysis, 4th edition, Cambridge University Press, reprinted 1963.
[WBI] Wikipedia, “Borwein Integral”,
en.wikipedia.org/wiki/Borwein_integral

[end]

PPRH-20201021 Page 21 21-Oct-2020

Page 85
2/24/2020 Władysław Narkiewicz – Wikipedia, wolna encyklopedia

Władysław Narkiewicz
Władysław Narkiewicz (born February 19, 1936 ) - Polish mathematician, known for his work Władysław Narkiewicz
on algebraic number theory , algebra and history of mathematics. Full professor since 1974.

Table of Contents
Biography
Some work
Footnotes
External links

Biography Country of Action Poland

He defended his doctoral thesis in 1961 and habilitated in 1967 at the University of Wrocław , Date of birth February 19, 1936
where he lectured from 1974 to 2006. He served in his life many different functions at the Professor of mathematical sciences
University. head of the Institute of Mathematics , dean of the Faculty of Mathematics and Physics Specialty: number theory, algebra,
and vice-rector for scientific matters [2] . history of mathematics [1]

Alma mater University of

In 1968 he was awarded the Stefan Banach [3] Wroclaw
Doctorate 1961
University of Wrocław
Some work Habilitation 1967
University of Wrocław
The Development of prime number theory . Springer , 2002.
Professorship 1974
Elementary and Analytic Theory of Algebraic Numbers . Ed. 3. Springer, 2004. ISBN 83-01-
13604-9 . Academic teacher
Number Theory . Ed. 3. PWN Scientific Publishing House , 2003. College University of
Wroclaw

https://pl.wikipedia.org/wiki/Władysław_Narkiewicz Page
1/2 86
2/24/2020 Władysław Narkiewicz – Wikipedia, wolna encyklopedia

Uniform distribution of sequences of integers in residue classes . Springer, 1984. ISBN 3- Employment 1974-2006
period
540-13872-2 .
Polynomial mappings . Springer, 1995. ISBN 3-540-59435-3 .

Footnotes
1. Home page - Władysław Narkiewicz (http://www.math.uni.wroc.pl/~narkiew/) , www.math.uni.wroc.pl [access 2017-11-27] .
2. Akademisches Kaleidoscope, page 6 (http://www.kaleidoskop.uni.wroc.pl/2006/15.pdf) (pdf)
3. Information on ptm.org.pl (http://www.ptm.org.pl/kategorie/konkursy/nagrody-glowne-ptm/nagroda-glowna-ptm-im-stefana-banacha?
page=11)

External links
Home page of Władysław Narkiewicz (http://www.math.uni.wroc.pl/~narkiew/)
[1] (http://genealogy.math.ndsu.nodak.edu/id.php?id=133245)
Popular (http://www.deltami.edu.pl/delta/autorzy/wladyslaw_narkiewicz/) - science articles (http://www.deltami.edu.pl/delta/autorzy/w
ladyslaw_narkiewicz/) in the Delta monthly

Źródło: „https://pl.wikipedia.org/w/index.php?title=Władysław_Narkiewicz&oldid=51831758”

This page was last edited Jan 20, 2018, 3:08 am. Text made available under a Creative Commons license: attribution, under the same conditions
(http:https://creativecommons.org/licenses/by-sa/3.0/deed.pl) , with additional restrictions. See detailed information on the terms of use
(http:https://foundation.wikimedia.org/wiki/Warunki_korzystania) .

https://pl.wikipedia.org/wiki/Władysław_Narkiewicz Page
2/2 87
---------- Forwarded message ---------
From: "Władysław Narkiewicz" <Wladyslaw.Narkiewicz@math.uni.wroc.pl>
Date: Sat, Feb 22, 2020 at 5:45 PM
Subject: Re: Open Letter Requesting Peer Review of Scientific Claim
To: P. Narasimha Reddy <nrriemann@sreenidhi.edu.in>

Dear Dr Reddy,

Thank you for sending me the information about the paper of Dr Eswaran
concerning Riemann Hypothesis. I had a short look at the paper and wrote
some small comments on it, which you will find in the attachment. I
observed that some introductory steps can have rather simpler proofs, but
noticed also that a part of the argument is unconvincing.

Yours sincerely

Wladyslaw Narkiewicz

SEE ATTACHMENT: From_Wladyslaw_22_Feb_2020_RHComment.pdf

Page 88
 In sections I and II I present two standard proofs of two assertions established in the paper, and in
sections III, IV I point out some incorrect steps.
I. Re section (4) of Extended Abstract
If λ(n) = (−1)Ω(n) is the function of Liouville, A is the set of all n ≥ 1 with λ(n) = 1 and

A(x) = #{n ≤ x : n ∈ A},

then
x
A(x) = + o(x). (1)
2

Proof. Since
∞
X λ(n) ζ(2s)
s
=
n=1
n ζ(s)

and 1 + λ(n) vanishes if λ(n) = −1 and equals 2 otherwise, we can write for <s > 1
∞  
X 1 X 1 + λ(n) 1 ζ(2s) 1 1
= = ζ(s) + = + g(s)
n ns n=1
2n s 2 ζ(s) 2 s−1
λ(n)=1

where g(s) is a function regular for <s ≥ 1. It remains to observe that now one can apply Ikehara’s theorem
to obtain (1).

II. Re section (5) of Extended Abstract

The sequence of values of the Liouville function λ(n) is not periodic.
Proof. Assume that for sufficiently large n, say for n > n0 the sequence λ(n) is periodic with period
N , thus

λ(n + N ) = λ(n) (2)

holds for n > n0 .
Choose an integer a > n0 with λ(a) = 1. The equality (2) implies that if b > a and b ≡ a mod N , then
λ(b) = 1, hence b is not a prime, thus any prime congruent to a mod N does not exceed n0 , and Dirichlet’s
theorem shows that this is possible only if GCD(N, a) > 1. Therefore if for an integer c > n0 one has
GCD(c, N ) = 1, then λ(c) = −1. If c1 , c2 both exceed n0 and are co-prime to N , then their product c1 c2 is
also co-prime to N , hence
−1 = λ(c1 c2 ) = λ(c1 )λ(c2 ) = (−1)(−1) = 1,
a contradiction.
III. Re section (6) of Extended Abstract
At the bottom of p.3 we read:
”Specifically, non-cyclicity would preclude any dependence of the type

λn = f (λn−1 , λn−2 , λn−3 , . . . , λn−M )

for finite M .”
This is simply untrue, as the example λ(6) = λ(2) · λ(3) shows. Since the function λ(n) is completely
multiplicative, every value of it at composite n depends on the smaller values of the function λ.
IV. Re Chapters 3–5
The author presents an argument to show that after a suitable permutation σ of summands in the series
P∞ P∞ h(n)
F (s) = n=1 λ(n)
ns one obtains a series Gσ (s) = n=1 ns converging in a region to the left of the line
<s = 1. The value of the sum of this series at a given point s0 depends on the permutation σ, and a theorem
of Riemann-Steinitz shows that if the series of F (s) does not converge at s0 and s0 is not a pole of F (s), then

Page 89
there are infinitely many complex numbers z such that with a certain permutation σ one has GP σ (s0 ) = z.
This shows that Gσ (s) mayP have no connection to the function F (s) = ζ(2s)/ζ(s) and the sum n≤x h(n)
is not related to the sum n≤x λ(n).
Moreover the uniqueness of expansion of a function into a Dirichlet series implies that both equations
(5.19) and (5.20) can hold only if one has λ(n) = h(n) for every n, thus the permutation σ is the identity.

Page 90
--

Dr. Kumar
Eswaran <kumar.e@gmail.com
>

to: Władysław Narkiewicz

<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

cc: "P. Narasimha Reddy"

<nrriemann@sreenidhi.edu.in>

date: Feb 24, 2020, 8:46 PM

subject: Response to Your Comments on the paper
on RH
mailed-by: gmail.com
Dear Dr Wladyslaw Narkiewicz,

I thank you for your thoughtful comments on the above subject, which has been forwarded to me
by the Convenor for my response.
My reply to your comments and clarifications are attached.

I thank you once again for the trouble you have taken to read my paper.

With Best Wishes

Regards
Dr K. Eswaran
Professor

SEE ATTACHMENT: Feb_24_2020_Reply_one_to_Wladslaw.pdf

Page 91
 Reply to Dr. Wladyslaw Narkiewicz

By K. Eswaran

February 24, 2020

Dear Dr. Wladyslaw Narkiewicz,

I thank you very much for your email and the trouble you have taken to read
my paper.
Regarding your comments (I) and (II):
I very much appreciate your eort to nd alternative proofs, it speaks a lot
for the genuine interest that you have in Mathematics, I applaud your eorts!
However, I have to study them very carefully, in order to see how they can
be used to prove RH.
Now I deal with your objection :(III)
I had made a statement 'Specically the non-cyclicity would preclude any
dependency of the type:
λn = f (λn−1 , λn−2 , λn−3 ,....,λn−M )
for nite M'
All I wanted to show is that the λ-sequence, is un-predictable. That is for
very large n, it is impossible to nd a xed integer M such that if you are given
the M consecutive values: λ(n − 1), λ(n − 2), λ(n − 3),...., λ(n − M ) (and no
other information) it is impossible for you to predict what the next value λ(n),
will be i.e. we will not be able to say if λ(n) = +1 or if λ(n) = −1. Just like the
situation when you have made very many coin tosses and you actually know the
values of M consecutive coin tosses say c(n − 1), c(n − 2), c(n − 3),...., c(n − M ),
but this information will not enable you to predict the next coin toss i.e. you
cannot tell if c(n) = +1(H) or if c(n) = −1(T). (H≡Head, T≡Tail). I follow the
method of proof adopted by Godel, the argument is as follows: If any number
belonging to a sequence is predictable, from a set of M previous values (M has
to be a xed number, however large), then there exists a recursive function such
as f (λn−1 , λn−2 , λn−3 ,...., λn−M ) which makes the prediction possible. If there
is no such xed M and function f then the number is unpredictable.
I think the above paragraph should clarify matters. Remember I want to
show that for very large N the function L(N) behaves like a random walk.(Note
I do not say that the λ-sequence is exactly equal to that of a coin toss (it is
certainly not), but I show that for large N the behavior of L(N) is like a random
walk and therefore the conditions of Littlewood's theorem are satised and RH
is therefore proved.

Page 92
 Your second objection (IV) is really the problem of notation.
To clarify I use the form (5.21) just to prove Littwood's theorem. This is
done from Equations (5.22) to (5.24), thus proving Littlewoods Theorem stated
in page 11, in the paragraph following (5.24). In actuality in order to study
L(N) I use only the form when λ(n) is substituted for g(n)or h(n).This is again
claried in the last two paragraphs after (5.26b) in page 12.
In brief I use only λ(n) to estimate L(N).
In summary, the proof of RH follows the 4 steps indicated in Ref[2], which
may be followed for guidance.
-x-x- X-
I hope I have claried all your points 1 . I will be most happy to reply to any
further points. Once again I thank you for your interest.
Regards
K.Eswaran (24 February, 2020)

1 You may wonder why I introduce Eq (1.2) in the form (3.10), when in actuality both
the forms are equivalent because each term which occurs in (1.2) also occurs only once in
(3.10).(in fact (3.10) the reordering of terms seems to make it unnecessarily complicated!).
This is because, when I discovered Eq (3.10), I quickly realized that the L(N) behaves like a
random walk. An examination of (3.10) led me to Figure 1 and the concept of towers,
which in turn lead me to the proof of RH. As stated in the second Footnote in page 6,
large part of section 3 and 4 and some of 5 can be omitted in a rst reading

Page 93
 - -----

"Władysław
Narkiewicz" <Wladyslaw.Narkiewicz@math.uni.wroc.pl
>
to: "Dr. Kumar Eswaran"
<kumar.e@gmail.com>

date: Mar 2, 2020, 1:34 PM

Re: Response to Your Comments

Dear Dr. Eswaran,

Thank you for your clarifications. I analyzed profoundly your paper in

which you describe the used arguments. I did not find the proof being
complete and my comments to it you will find in the attached file.

With my best wishes

Wladyslaw Narkiewicz

SEE ATTACHMENT: From_Wladyslaw_2nd_March_2020_Further_Comments.pdf

--

Page 94
 Further comments
1. Your main idea is to use the observation that the function of Liouville resembles the coin-tossing
sequence. This observation is correct, the sequence λ(n) can be regarded as a realization of the random
sequence of numbers 1 and -1, having both the probability 1/2. Based on this observation one can expect
that for the sum
N
X
L(N ) = λ(n)
n=1

the inequality
|L(N ) ≤ c(ε)N 1/2+ε (1)
holds for every ε > 0, showing the convergence of the series
∞
X λ(n) ζ(2s)
f (s) = =
ns ζ(s)
k=1

in the open half-plane <s > 1/2, hence implying the regularity of ζ(2s) ζ(s) in that half-plane. Riemann’s
Hypothesis would be an obvious consequence.
One has only remember all assertions about infinite random sequences (in particular for the coin-tossing
sequences) hold with probability 1, i.e., not for all sequences but only for almost all such sequences. Hence
if one studies a particular realization, then each needed property of random sequences must get a proof.
Arguments per analogiam are not permitted.
Similarity of some properties of the sequence λ(n) to properties of random sequences has been studied
in the last years by several authors. Usually one says that λ(n) is a pseudo-random sequence. There are
several open questions concerning this similarity. For example, if a(n) is a random sequence consisting of
numbers ±1, then for every k there is a positive probability for the set of numbers m such that

a(m + j) = εj

happens for j = 1, 2, . . . , k, where εj ∈ {−1.1} is given. It has been conjectured by S.Chowla in 1965 that
also the sequence λ(n) has this property, but this has been established only for k = 3 by A.Hildebrand in
1986 (Math.Proc.Cambridge Math. Soc. (1986), 229–236), and the probabilities in this case were recently
determined by K.Matmäki and M.Radziwill (Forum Math. Sigma 4 (2016), e14, 1–44).
This indicates that not every property of the coin-tossing sequence √ transfers automatically to the se-
quence λ(n), and this applies in particular to the assertion |L(N )| ≤ c N , used in your proof of RH, which
is an analogue of the true corresponding assertion in the case of coin-tossing.
A similar attack on the Riemann Hypothesis using the Moebius function µ(n) has been tried by
T.J.Stieltjes (Comptes Rendus Acad. Sci. Paris 10 (1985), 153–154) in 1885, who asserted the conver-
gence of the series
∞
X µ(n)
n=1
nx

for all x > 1/2. He did not publish a proof and nothing about it has been found in his preserved notes, but
in a letter to E.Hermite he wrote that the proof is based on the inequality
N 
X  √
|M (N )| =  µ(n) ≤ B N
 
 
n=1

with a suitable constant B.

2. On page 2 of your paper you want to show that the sequence λ(n) behaves like a sequence of coin
tosses and state two its properties, equal probabilities and non-periodicity and independence.

Page 95
 a) Equal Probabilities.
I understand that you mean by that the following correct equalities concerning A+ (N ) = #{n ≤ N :
λ(n) = 1} and A+ (N ) = #{n ≤ N : λ(n) = −1}:

A+ (N ) A− (N ) 1
lim = lim = . (2)
N →∞ N N →∞ N 2
This is actually an assertion about the density of the sets A+ (N ) and A− (N ) and to be able to interpret
it in probabilistic language one has to define the underlying probability space. Unfortunately, there is not
possible to define a probability space on the set of positive integers in which every number has the same
probability, and to go around this trouble one usually considers probability on the set of the first N positive
integers, with every its element acquiring probability 1/N . Then one applies the existing probabilistic tools
to study the considered problem, and looks what happens when N tends to infinity. Sometimes this leads
to good results (see e.g. the two-volume book by Peter Elliott [”Probabilistic Number Theory”, Springer
1979]).
You write on p.23 that the equalities (2) are consequences of Theorem 1 on that page, but this is
incorrect. Theorem 1 states only that there exists a bi-unique correspondence A+ ⇐⇒ A− between the sets
A+ = #{n : λ(n) = 1} and A− = #{n : λ(n) = −1}, hence they have the same cardinality, but this
happens for all pairs of infinite subsets of positive integers, and if this theorem would imply (2), then it
would also imply the following completely false equalities:
If π(N ) denotes the number of primes p ≤ N , and X(N ) denotes the number of composite numbers
≤ N , then

π(N ) X(N ) 1
lim = lim = ,
N →∞ N N →∞ N 2
whereas the first limit actually equals 0 and the second equals 1.
The equalities (2) are not new, being equivalent to
1 1 1
lim #{n ≤ x : 2|Ω(n)} = lim #{n ≤ x : 2 - Ω(n)} = , (3)
N →∞ N N →∞ N 2
whose proof has been indicated already in 1898 by H. von Mangoldt, and details were given in §167 of the
book of E.Landau (”Handbuch der Lehre von der Verteilung der Primzahlen”), published in 1909. Another
proof, based on a tauberian theorem of Ikehara, I showed you in my previous message.
b) Non-periodicity and Independence
Your analytical proof given in Appendix III that the sequence λ(n) is non-cyclic is correct.
There is a problem with the notion of independence. On p.3 you define the dependence of values of the
function λ writing: ”. . . λ(n) is dependent on λ(m), if the latter is required to find the former ”, but this has
no mathematical meaning. Here λ(n) is a well-defined function for every positive integer n and the whole
sequence of its values at integers is well-defined from the very beginning. This makes the situation essentially
differing from a coin-tossing sequence whose elements are created consecutively.
On the bottom of p. 2 you wrote ”Specifically, non-cyclicity would preclude any dependence of the type

λn = f (λn−1 , λn−2 , λn−3 , . . . , λn−M ) (4)

for finite M .”
This is not correct, just look at the Fibonacci sequence defined by F1 = F2 = 1, Fn+1 = Fn + Fn−1 . It
tends to infinity, hence is non-cyclic, but satisfies a relation of type (4) with f (X, Y ) = X +Y . Non-existence
of a relation of type (4) for a particular function, say X(n), shows only that the sequence X(1), X(2), . . . is
not a recurrence sequence.
You use later in Appendix IV your assertions discussed above in a) and b) to claim that L(N ) series is a
random walk and deduce in Theorem 4 the bound on L(N ) needed in the proof of RH, but as I noted above,
the assertion b) has no correct proof. The discussion of independence in Appendix IV lacks mathematical
precision and the reader can wonder how this discussion implies that the sequence λ(n) is equivalent (what
kind of equivalence is here used?) to coin tosses, as indicated in the title of the Appendix.

Page 96
--

Dr. Kumar
Eswaran <kumar.e@gmail.com
>

to: Władysław Narkiewicz

<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

date: Mar 5, 2020, 7:48 AM

subject: My Response to your Further Comments
mailed-by: gmail.com
Dear Professor Wladyslaw Narkiewicz,

I thank you very much for your detailed reading and comments. I am
truly grateful for the time that you are spending and I am very much
indebted to you for this!

Please find my response in the first attachment.

Thank You Once again.
Regards
K. Eswaran
P.S. This email contains two other attachments, for your convenience.

SEE ATTACHMENTS: (1) March_5_2020_reply_Two_Wladyslaw.pdf

(2) Pathway_to_RH_lecture_by_Eswaran_IIT_Madras.pdf

(3) Summary_of_lecture_IIT_Madras.pdf

------

Page 97
 Dear Professor Wladyslaw Narkiewicz,
I thank you very much for your detailed reading and comments. I am truly
grateful for the time that you are spending and I am very much indebted to you
for this!
Firstly, I reply to the points you have made in page 2 that are of immediate
relevance to my proof of RH.

(a) Equal Probabilities

Since by denition λ(n) = (−1)Ω(n) , where Ω(n) is the number of prime
factors of n, multiplicities included, with the denition λ(1) = 1. Hence, λ(n) =
−1,not only when n is a prime number but also if n is a composite number with
odd number of prime factors, e.g. 12= 2x2x3, hence λ(12) = −1. If we denote
the number of composite integers n less than or equal to N having odd number of
prime factors i.e.( λ(n) = −1),as X− (N ) and the number of composite integers
n less than or equal to N having even number of prime factors i.e.( λ(n) = 1), as
X+ (N ). Therefore, the equality that you wrote down should actually read as:

π(N ) + X− (N ) X+ (N ) 1
lim(N → ∞) = lim(N → ∞) =
N N 2
there is no contradiction.
In the proof of equal probabilities, I purposely avoid falling into the trap of
using Cantor's mapping technique. I label every integer by a unique triad of
integers. Then I sort all the integers into sets called Towers', each integer in
a Tower is followed by an unique (higher) integer in the same tower and their
λ-values alternate as + 1 then -1, then +1 etc. Further, every integer occurs
only once in some tower or other and I have further ensured that for the special
integers say m, which occur at the base of any Tower, I have arranged matters
such that if for every integer m at the base which has λ(m) = 1 there is another
unique integer m0 at the base of another Tower which has λ(m0 ) = −1 and
vice-versa. All these considerations, in my opinion, make the proof of equal
probabilities for (N → ∞) water tight.
I have given another Alternative Proof for equal probailities. The only as-
sumption made in this second proof is that in the set of all ordered integers
every odd integer is followed by an even integer and precedes another odd in-
teger -thus making the number of even integers equal to the number of odd
integers. The second proof is by construction and therefore avoids mappings.
The link to the Alternative proof is given as Ref[3) in the rst Invitation Email
Sent to you.

(b) Non-periodicity and Independence

The Fibonacci series is not the right analogy because every term in the series
takes newer and newer integer values and there is almost no repetitions except
for the 2nd and 3rd terms. But the λ− series is extremely constrained, each
term can take either +1 or −1.

Page 98
 It is because of this reason that for some large N0 all the λ(n), (with n > N0 ),
cannot be predicted by any formula of the form:

λ(n) = f (λ(n − 1), λ(n − 2), λ(n − 3), ...., λ(n − M ))

where M, is a xed integer, this is because the function f takes as inputs
M consecutive values of λ0 s and evaluates to +1 or -1 to predict the next λ viz
λ(n). But these M consecutive values of λ0 s can be thought of as a string of M
numbers, but since each number can take only the values +1 or -1, there can
be only 2M dierent patterns of strings each of length M. Hence this means
the above formula will at most predict a sequence of λ0 s which has a maximum
M
length of L=2 and after this the entire predictions will repeat and thus form
a cycle. [QED}
I enclose as an attachment to this Email,(File called ESWARAN LG FULL
LECTURE) my Invited Lecture to IIT Madras which gives more examples please
see slide 57. The other slides of the lecture can also be consulted for the deriva-
tion of Equal Probabilities and other matters, including Alternative Proofs (see
slides 69-70)of several theorems I have proved in the Main paper.

My General Comments and Closure

The Extended Abstract Ref[2] in the rst Email describes the crucial Steps
used to provide the proof of RH. For your convenience, I have attached this as
a le called SUMMARY of LECTURE IIT

The basic idea in the proof is to show that for very large N (actually at inn-
ity), L(N) behaves like a random walk. This is the requirement laid out by Lit-
tlewood's Theorem to prove RH armatively. Now we know that the λ−series
is deterministic and of innite length, so it strictly is not exactly obtainable
by coin tosses. However, for very large N it so happens that the deterministic
nature of the λ0 s do not eect the essential statistical similarities between the
λ−series and a series of coin tosses. In a random walk it has been proved that
√
the root mean square distance in N steps has a square root behavior (viz N)
for large N. The derivation for this expression has been done (example see pages
3-5, in Stochastic Problems in Physics and Astronomy by S. Chandrasekar Rev.
of Modern Physics (1943) vol. 15, no. 1), by assuming that each step (like coin
tosses) have (i) Equal probabilities of being (+1) or (-1), (ii) Independence (iii)
unpredictability. So any sequence which has all these properties must have the
same square root behavior for their root mean square value, it so turns out that
for large N the λ−series has all these three properties and therefore L(N ) will
also have the square-root behavior, proving RH.
I hope I have claried the points you have raised.
I will be ever very grateful for the time you have taken. It has
been a pleasure to read and think about your insightful comments.
Regards K. Eswaran 5th March 2020

Page 99
------

"Władysław
Narkiewicz" <Wladyslaw.Narkiewicz@math.uni.wroc.pl
>
to: "Dr. Kumar Eswaran"
<kumar.e@gmail.com>

date: Mar 9, 2020, 4:51 PM

subject: Re: My Response to
your Further
Comments
Dear Professor Eswaran,

I looked again thoroughly at your text and your explications. I subsumed

my thoughts about your paper in the attached file. Unfortunately there is
a deep hole in your argumentation, when you assert without any proof that
a theorem in the theory of random sequences implies a bound for the sums
of values of the function $\lambda(n)$.

Do not worry about this situation. Several excellent mathematicians tried

without success to prove Riemann Hypothesis.

With every good wish

Wladyslaw Narkiewicz

SEE ATTACHMENT: From_Wladyslaw_9th_March_2020_Final_Comments.pdf

Page 100
 1. The central point in your method is the proof of the inequality
 
XN 
|L(N )| =  λ(n) ≤ BN 1/2+ε (1)
 
 
n=1

for every positive ε, with the constant B depending on ε. This is actually an easy observation that (1) implies
Riemann Hypothesis, because by elementary partial summation one deduces from (1) the convergence of the
series
∞
X λ(n)
f (s) = (2)
n=1
nx

for all real x > 1/2, and by the theory of Dirichlet series this implies the convergence of that series in the
open half-plane <s > 1/2 to a regular function. Since for <s > 1 the sum of this series equals ζ(2s)/ζ(s), we
P in <s > 1/2,
get f (s) = ζ(2s)/ζ(s) also in that half-plane and it follows that ζ(2s)/ζ(s) does not have poles
hence Riemann Hypothesis follows. The same argument works also for the sum M (N ) = n≤N µ(n) and
in that case occurs already in the failed attempt by Stieltjes.
The main problem with your approach to RH lies in the proof of (1).
2. At the end of your last message you wrote that any sequence which has the following properties:
(i) Equal probabilities of being 1 or −1,
(ii) Independence,
(iii) Unpredictability,
must have the same behavior as a random walk, and applying this to the sequence λ(n) you infer the
inequality (1).
On your slide 57 you define the notion of independence, which shows that you call elements of a sequence
a1 , a2 , . . . independent if the sequence is not a recurrent sequence. Unfortunately you do not give a precise
definition of unpredictability. Mathematics is a formal science, and every notion occurring in a mathematical
argument must have a proper formal definition. I understand that you use the word ”unpredictable” in the
common sense, but this makes your argumentation incomplete.
I did not find in your paper a proof of the assertion that the conditions (i), (ii) and (iii) (whatever they
mean) imply for the sequence λ(n) the inequality (1). I presume that you had in mind an application of the
following well-known theorem about simple random walk:
of independent random variables with values ±1,
THEOREM. Let X1 , X2 , . . . be an infinite sequence P
N
each of probability 1/2. If SN denotes the random variable j=1 Xj , and E(|SN |) denotes the expected value
of the random variable |SN |, then the limit

E(|SN |)
lim
N →∞ N
p
exists and is equal to 2/π = 0.7978 . . . .
Unfortunately, nowhere in your paper you mention how one can formally deduce your assertion about
(1) from this well-known theorem. I wonder whether such a proof is possible, as the elements of the sequence
λ(n) depend on n, and elements in a random sequence do not have that property.
This makes it clear that the presented proof of Riemann Hypothesis is not complete.
3. In my previous message I observed that your way of establishing the ”equal probabilities” property
for the sequence of values of the λ-function can lead to false result. You answered that the equality which
I wrote in that message should look in another way, avoiding contradiction. It seems that I presented my
argument in a not sufficiently precise way. This time I will try to do better.
Let P denote the set of primes arranged as a sequence 2 = p1 < p2 < p3 · · ·, and let C : 1 = c1 < 4 =
c2 < c3 < · · · be the set of all remaining positive integers, also arrangd in a sequence. Now I apply your own
argument from your last message to obtain the proof that the conditions ”an integer n lies in P ” and ”an
integer n lies in C” have ”equal probabilities”. Note that every positive integer lies either in P or in C and

Page 101
the sets P, C have no elements in common. With every prime pn I associate the unique number cn lying in
C. And now I copy your argument in this case:
”I have arranged matters such that for every integer m in P there is another unique integer m0 in C
and vice-versa. This makes the proof of equal probabilities water tight.”
Now observe what happens. Denote by π(N ) the number of primes pn ≤ N , and by C(N ) the number
of numbers cn ≤ N .
The equal probabilities property implies

π(N ) c(N ) 1
lim = lim = ,
N →∞ N N →∞ N 2
but this is not true, as actually one has

π(N )
lim = 0.
N →∞ N

All this is of minor importance, as the result which you wanted to prove is true, found by von Mangoldt
already in the 19th century.

Page 102
On Fri, Mar 13, 2020 at 7:57
AM Dr. Kumar Eswaran
<kumar.e@gmail.com>
wrote:
Dear Wladyslaw Narkiewicz,

I thank you very much for

your detailed reading and
comments. It is my very
great good fortune to have a
learned person like you to be
reading my papers.

I agree with you that I must

give proper Definitions in a
Mathematical paper.

I am following your advise.

At the moment, I am in the
process of framing my
arguments and basing them
on properly defined entities
and terminologies.

This is taking me some time.

I will get back to you in 5 or 6
days.

Regards
K.Eswaran
---------

Dr. Kumar
Eswaran <kumar.e@gmail.com
>

to: Władysław Narkiewicz

<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

date: Mar 17, 2020, 2:29 PM

subject: My Reply to your latest comments of March
9th 2020
Dear Wladyslaw Narkiewicz,

Page 103
 I want to once again express my thanks to you for sparing so much of your time reading my
work. I do not know how to express my gratitude except by taking every word of yours seriously
(thus demonstrating my profound respect for you) and try to make a complete clarification to the
best of my ability.

Please find my detailed response to your queries in the attached.

I will be most happy to clear any further doubts (if any).

God Bless!
Regards
Kumar Eswaran

Please See Attachment,Thank You!

See ATTACHMENT: March 17-2020_Reply_three_to_Wladyslaw.pdf

-------

Page 104
 Letter to Prof
Dear Professor Wladyislaw Narkiewicz,
0 Wladyslaw
I now dene predictability of the λs follows.
Narkiewicz
DEFINITION: We say that λ(n) is predictable if, there exists a nite integer
M , such that for every n > N0 , the λ(n) , is derivable from its M previous values
λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ).
Note, the value of n is not explicitly known: only the values {λ(n − r), r =
1, 2, ..M } are known.
We then assume that if λ(n) is not predictable then it is independent.
Of course, if λ(n), is derivable as above, then there exists some function f
s.t :

λ(n) = f (λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ))

The justication for these denitions are given in Section 1 and Section 2.

SECTION 1:
Regarding your objection to the treatment of Eq(1) viz.

N
|L(N )| = Σn=1 λ(n) (1)

which by Littlewood's Theorem is supposed to satisfy the relationship:

|L(N ) ≤ C N a+ε (2)

for large N. In (2) I have used a in the exponent on the R.H.S. instead of
1/2, according to Littlewood if it so happens that a = 1/2. then RH is proved.

Comparing the above with the summation of a sequence coin tosses (or
random walk) depicted by:

N
|SN | = Σj=1 Xj (3)

where the random variable Xj takes on values +1 or −1, with equal prob-
ability and are independent of each other then, it is well known that for large
N:

2√
r
|SN | ≈ N (4)
π
Now, we have already shown that the λ(n) has the probability of being +1
or −1 for very large n. In addition I have proved (in Appendix IV) that for
large n, there is no way of knowing λ(n + 1) λ(n).1 This
if we happen to know
was sucient to prove independence for large n. (Note if we know λ(n) the
next predictable value is λ(2n) = −λ(n) , and 2n is very far away from from n
for large n.) Later, I used another criterion of independence surmising that if

1 In fact, in Appendix IV, we have an alternative proof of 'independence' because we showed

that the sequence of λ0 s between λ(n + 1) and λ(2n) are independent of each other; this length
of this strip becomes innite as n tends to innity.

Page 105
λ(n+1) is derivable (predictable) from its M previous values, then the λ(n)0 s are
predictable and hence not independent. But I showed that no xed nite M ex-
ists
2 and hence we can consider the λ0 s are independent for large n. (Therefore,
this second criterion is more general because the rst corresponds to M=1). We
show in the second Section 2 that this denition of predictability is the correct
one because by proving the λ0 s are un-predictable (by this denition) and hence
0
independent is able to explain the behavior ('phenomena') of the λ s over large
consecutive sections.
Hence we see that by using the same logic which derives (4) from (3) we
can derive (2) from (1) and we will have the position of the critical vertical line
at a = 1/2 thus proving R.H. However, in the paper I use a yet more rigorous
analysis (see last paragraph of page 14 of the main paper) which was done in a
study on independent random variables done by Khechin and Kolmogorov using
iterative logariths, their analysis gives the dependence of ε on N and how as N
tends to innity. From their result I have shown that the width of the critical
line tends to zero, proving that all the nontrivial poles of F(s) and hence the
zeros of ζ(s) lie on the critical line. QED

has

SECTION 2 : Phenomenological study of behavior of λ(n)0 s

In Section2 of Appendix VI of the Main paper, I have calculated consec-
utive λ(n)0 s forming a large sequences of length M of the form Λ[N0 , M ] ≡
{λ(N0 ), λ(N0 + 1), λ(N0 + 2), ..., λ(N0 + M − 1)}.where N0 is some large in-
2
teger. And for each such case I did a χ t which compare sit with a Bino-
mial sequence (coin tosses) of the same length M. as described by Knuth . In
each case the value of χ2 ≤ 4.0 showing that the λ0 s in the sequence are in-
distinguishable from coin tosses. In the Tables given in the Appendix VI , I
√
typically choose N0 = N (a perfect square andM = N . E.G. the items in
row 2 Table 1.4 show that the sequence of length M = 100, 000 starting from
λ(10, 000, 000, 001),......,to λ(10, 000, 100, 000) has a χ2 value equal to 1.15,thus
it is statistically indistinguishable from a sequence of 100, 000 coin tosses or a
random walk of 100, 000steps. This phenomena' depicting the statistical be-
0
havior of the λ s happens for all the sequences of ALL the entries in the Tables
in Appendix VI. The reason for this phenomena is because each λ in the set
√
Λ(N, N ) is sampled
√ from a dierent 'Tower' and computing the summatory
function L(N, N ) is like randomly picking a number from the next Tower
which then may have value +1(H) or -1(T) with equal probability.
3
Since all the sequences Λ[N0 , M ] , of diering values of N0 and M , behave
like coin tosses regardless of their starting value N0 , we can say that the statis-
tical behavior does not depend on the starting value N0 ,further the statistics of

2 If
no xed nite M exists (i.e. M grows large with n or is innite) it automatically
means that the λ(n + 1) is all the more unpredictable from λ(n) thus the λ0 s are practically
independent for large n.
3 It is possible to argue similarly, and demonstrate that the sequence Λ(N , M ), for large
0
N0 and M < 2N0 will also statistically behave like coin tosses because each λ in the sequence
will be drawn from a dierent Tower. and could be randomly +1(H) or -1(T).

Page 106
Λ[N0 + k, M ] (k being a small number) is the same as that of Λ[N0 , M ] we call
this property as Translational Invariance which is certainly true if we are talk-
ing of the N0th coin toss c(N0 )(≡ XN0 ); the coin tosses are all unpredictable and
therefore independent. Now the numerous computed χ2 values show that the
0
sequences of the λ s are statistically indistinguishable and also have the property
of being statistically Translational Invariant for their corresponding sequences
of coin tosses. Hence, to actually explain this phenomena of the statistical be-
havior we must show that the λ0 s are also statistically Translational Invariant,
'unpredictable' and therefore 'independent'. Hence we are by tour-de-force led
to the following denitions:
DEFINITION: We say that λ(n) is predictable if, there exists a nite integer
M , such that for every n > N0 , the λ(n) , is derivable from its M previous values
λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ).
Note, the value of n is not explicitly known only the values {λ(n − r), r =
1, 2, ..M } are known.
We then assume that if λ(n) is not predictable then it is independent.
Of course, if λ(n), is derivable as above then there exists some function f s.t
:

λ(n) = f (λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ))

In the Main Paper we proved that such a function cannot exist otherwise it
will make the λ-sequence cyclic, proving un-predictability and hence indepen-
dence of the λ0 s.
From the above analysis we can conclude that as N tends to innity the
statistical behavior of the λ0 s and the 'coin tosses' become very similar. Thus
just as Eq.(4) follows from Eq. (3) for coin tosses, we are forced to conclude
that Eq. (2) (with a=1/2) follows from Eq. (1) for the λ0 s .
4
We have thus shown that the λ-series has a dominant statistical behavior of
a random walk and therefore the summatory function L(N ) can be computed
as is done in the paper using the iterative logarithm formula derived by Khechin
and Kolmogorov, thus proving R.H.
It cannot be over emphasised that apart from proving RH, I have given
reasons for the phenomena of large subsets sets of consecutive
p √ p λ0 s denoted as
S+ (N ) = Λ(N + 1, N) and S− (N ) = Λ(N − N + 1, N ) , (N a square
integer) behaving like coin tosses, (see Eqs. (13) and (14), Sec. 4, Appendix VI
of Main paper). This is important because by taking a collection of all perfect
squares N, the sets S+ (N )and S− (N ) contain all the λ0 s upto innity. Since,

4 This is because the deterministic formula λ(p.n) = λ(p).λ(n) = −λ(n), which can be used
to predict the other values of the λ√0 s having known λ(n) does not disturb the statistics of any
large sequence, for example Λ[N, N ] (N being square integer), this is because even if λ(n)
belongs to the sequence, even the very next√predictable value λ(2n) lies outside the range
of the sequence and does not occur in Λ[N, N ]. The predictive power of the deterministic
formula for λ(m.n) = λ(m).λ(n) , causes only a minor perturbation on the dominant statistical
'random walk' behavior of L(N), for very large N, and whose eect fades away at innity.
In my project log in my Researchgate webpage, I have examined the behaviour of λ0 s and
its eect on L(N) and they have conrmed this assertion.

Page 107
the Tables given in Appendix VI are actual computed values of the λ0 s we have
not only proved RH but have given an explanation of the behaviour of the λ0 s .
Therefore, I believe that: Even if one is hard pressed to deny this proof of RH,
then he/she is compelled to deny the very existence of this phenomena, but the
latter cannot be denied because too many computations and χ2 comparisions
have been conrmed. It is because of these reasons (and of course my own
intution), I humbly believe that, what we have is a proof of RH!
In this reply I have conned myself to only your objections raised as points
1 and 2. I am leaving out point 3, as you have said a proof by von Mangoldt
already exists. (I have given (in Ref 3, in the 1st Invitation Letter (email) dated
3rd Feb 2020, an alternative proof which relies on induction and not mapping,
but I acknowledge that it has now become academic and redundant!).
To conclude, Professor Wladyslaw Narkiewicz, I want to once again express
my thanks to you for sparing so much of your time reading my work. I do
not know how to express my gratitude except by taking every word of yours
seriously (thus demonstrating my profound respect for you) and try to make a
complete clarication to the best of my ability. I will be most happy to clear
any further doubts (if any). God Bless!
Regards
Kumar Eswaran
March 17, 2020


Page 108
-------

"Władysław
Narkiewicz" <Wladyslaw.Narkiewicz@math.uni.wroc.pl
>
to: "Dr. Kumar Eswaran"
<kumar.e@gmail.com>

date: Mar 30, 2020, 8:08 PM

subject: Riemann Hypothesis
mailed-by: math.uni.wroc.pl
Dear Professor Eswaran,

In my previous messages I tried to show you that your arguments are not
very convincing. Your proof consists of the following steps:

1. You define predictability of a sequence in the following way:

The sequence $a_n$ is predictable if for some integer $M$ the term $a_n$
(for large $n$) depends on $M$ precedent terms.

This definition coincides with the definition of a recurrent sequence,

used in several parts of mathematics.

2. You state that a sequence which is not predictable is independent,

without defining the word "independent".

3. You recall a theorem from probability theory describing asymptotical

behavior of the expected value of a series of independent random
variables. I am not quite sure whether you quoted it correctly. I now it
only for the case when the random variables attain the values 0 and 1. In
the case when you have 1 and -1 it seems that the expected value would
tend to zero. But I am not an expert in probability.

4. You state that this theorem implies that if a sequence is

unpredictable (in the above sense) and attains the values 1 and -1 with
the same frequency, then the sum of its first $N$ values is asymptotic to
$c\sqrt N$ with some $c$.

5. Applying this to the sequence $\lambda(n)$ one obtains the Riemann

Hypothesis, as it is well-known that if for all large $N$ one has
$$\sum_{n=0}^N\lambda(n)\le c\sqrt N,\eqno(*)$$
then RH follows.

Page 109
You do did not a proof the step 4, and actually the assertion stated
there is incorrect, as one can easily construct a sequence $a_n$ with
$a_n=1 or -1, which is unpredictable in the sense of 1, attains its values
with the same frequency but does not satisfy the assertion (*).
Example:

For $k=1,2,\dots$ denote by $I_k$ the interval

$[(k-1)\cdot10^4,k\cdot10^4-1]$ and denote by $A_k$ the set of all even
integers in $I_k$ to which $k^{3/4}\cdot10^3$ odd integers from $I_k$,
taken at random, are added, and let $B_k$ be the set of all remaining
integers from $I_k$. Let $A$ be the union of all sets $A_k$ and $B$ the
union of all sets $B_k$, and define $A(x), B(x)$ to be the numbers o
elements of the sets $A,B$ in the interval $[1,x]$.

Define a function $f(n)$ by $f(n)=1$ for $n\in A$ and $f(n)=-1$ for $n\in
B$, and observe that because of the randomly added elements to $B$ the
function $f$ satisfies your unpredictability condition.

One sees immediately that we have

$$\lim_{x\to\infty}A(x)/x =\lim_{x\to\infty}B(x)/x = 1/2,$$

and
$$\sum_{n\le k\cdot10^4}f(n) = A(k\cdot10^4} - B(k\cdot10^4)) =
k^{3/4}\cdot10^3,\eqno(**)$$
but according to your assertion 4 this sum should be bounded by
$$c\sqrt{k\cdot10^4} = c\sqrt k\cdot10^2,$$
in contradiction to (**).

Therefore your proof is not correct.

Do not worry, your are not the first person who presented an incorrect
proof of RH. At their list one finds some excellent mathematicians.

With every good wish

Wladyslaw Narkiewicz
Dr. Kumar
Eswaran <kumar.e@gmail.com
>

Page 110
 to: Władysław Narkiewicz
<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

date: Apr 2, 2020, 8:34 PM

subject: Response to your Email of the 30th March
Re: Riemann Hypothesis
Dear Professor Wladyslaw Narkiewicz,

I thank you for your email of the 30th March; I wanted to immediately respond by
pointing out that you had overlooked a crucial criteria which needs to be satisfied by
your Example but it does not, thus invalidating it from being a valid contradiction.
However, since you have been taking so much of interest and trouble in studying my
work, I felt that a short immediate response will be discourteous and disrespectful from
my side. So I decided that I would write out in fair detail the concepts, the logical
sequence of thoughts and the sequence of findings which enabled me to come up with
this purported proof of RH. All this took time and therefore this delay in my response.

Yes, the proof involves many areas: statistics, randomness, complex functions,
concepts from computer science and information theory (Knuth, Shannon) that I
borrowed apart from number theory. A proper understanding of the proof and of the
methods employed would need a fair understanding of most of the above.

I plead for your patience in the perusal of the write-up (attached) and I hope it will be
able to clarify much of what I have done and it will be able to clear any misgiving that
you may have.

I will be ever grateful for your interest, Thank You!

Kind Regards

Kumar Eswaran

This Email is in response to your Email:

On Mon, Mar 30, 2020 at 8:08 PM "Władysław Narkiewicz"
<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

SEE ATTACHMENT: April_2_2020_Reply_Four_to_Wladyslaw.pdf

- ---------

Page 111
 MY WRITE UP IN RESPONSE TO YOUR EMAIL of 30th March
2020
2nd April 2020

Dear Professor Wladyslaw Narkiewicz,

I thank you for the trouble you are taking over my papers.
At the outset I wish to say that the summatory function L(N ) :

N
|L(N )| = | Σn=1 λ(n) | (1)

and the summation of a sequence coin tosses (or random walk) depicted
by:

N
|SN | = | Σj=1 Xj | (2)

(where the random variable Xj takes on values +1 or −1, with equal prob-
ability), will behave statistically identically for large N, if ALL the following
conditions are satised by λ(n) viz. (i) Equal probabilities, (ii) independence
If all these
and (iii) unpredictability (see Extended Abstract for details).
conditions are satised then only we can conclude that, for large N:

|L(N ) | ≤ C N 1/2 +ε (3)

P
However, in constructing your Example (refer to your para 5) for f (n),
you have overlooked one crucial requirement. The sequence f (n) which you con-
structed though you say is unpredictable has the property that if n is even then
not only f (n) = 1 but f (n + 2) = 1 so one cannot say f (n + 2) is not deducible
2
from f (n). The sequence f (n) will invariably fail the χ test because most of
the times f (n) and f (n + 2) are highly correlated; the test of independence will
fail. All the conditions (i), (ii) and (iii) must be satised, but this has not hap-
pened and because of this your result is not a valid counter example. I repeat
any sequence to resemble a sequence of coin tosses MUST satisy all the three
criteria: (i), (ii) and (iii). In other words, for a valid comparision, one cannot
take any arbitrary sequence, of +1's and -1's, because one needs to show equal
probability and unpredictability and independence before one can even begin
to consider the sequence as an appropriate counter example. The reference of
D. Knuth gives details of how such χ2 tests are done. In the write-up below, I
have described the results of extensive and very many computations over large
streches of consecutive λ0 s the χ2 test has always passed, thus placing the matter
beyond reasonable dispute.

Before answering your other queries, I wish to say that if there is a random
variable Xj which takes on values +1 or −1, with equal probability and is
independent i.e. they satisy the three criterion: viz. (i) Equal probabilities, (ii)
independence and (iii) unpredictability then, it can be proved that, for large N,

2√
r
|SN | ≈ N (4)
π

Page 112
.The proof of Eq4. , following from the denitions (2) and the three criteria (i),
(ii) and (iii), is very well known result in the Statistical theory of random vari-
ables pertaining to coin tosses or 1-d random walks.See references in Footnote:
1
Therefore, any other sequence consisting of numbers +1 or -1,
which satises the above three criteria (i), (ii) and (iii), will have
its sum of N terms behave like (4). For convenience I call this as
the Principle of Statistical Analogy. 2 Hence, by this reasoning Eq. (3)
will follow from Eq. (1), because we have proved that the λ(n) satises the
three criteria (i), (ii) and (iii). Thus RH necessarily follows from Littlewood's
theorem. Additional Note: Littlewood's Theorem only requires that the three
criteria need only be satised by the λ-sequence for large N and we have proved
that this is so.
Coming to your other queries: In my previous email I gave reasons for the
natural denitions of un-predictability, independence, non cyclicity, and Equal
probabiliy. For the latter situation viz equal probability, I dene probability in
the somewhat common sense way that is: given a randomly chosen integer
n,then λ(n) has an equal probability of having a value +1 or -1. Since λ(n) is
dened over all integers and can have only two possible values (+1 or -1) there
is no need to build an elaborate machinery of probability spaces, Borel sets,
Lebesgue measures etc.
All the denitions adopted by me has the virtue of being natural denitions
and simple and sucient for my purposes: which was to show that the three
criteria is satised by the λ-sequence.

-
Dear Professor, As far as I can tell, I have replied to almost all your queries.
Now, I wish to explain in my own way of thinking, my surmise that the RH
is true. In order that you will appreciate what I say I now request you to think
from my point of view i.e. the point of view of a Theoretical Physicist,
3 which
is some what dierent form the point of view of a strict Number Theorist as I
will presently explain. I request you to read the rest of this email and try to
empathise with my point of view.
In mathematics a Mathematician starts from some initial premises which
are axioms e.g. a geometer starts from (say) Euclid's Axioms and a Number

1 See e.g. (1) Chandrasekhar S., (1943)'Stochastic Problems in Physics and Astron-
omy',Rev.of Modern Phys. vol 15, no 1, pp1-87,
(2) Khinchine, A. (1924) \Uber einen Satz der Wahrscheinlichkeitsrechnung", Fundamenta
Mathematicae 6, pp. 9-20,
(3) Kolmogorov, A., (1929) \Uber das Gesetz des iterierten Logarithmus". Mathema-
tische Annalen, 101: 126-135. (At the DigitalisierungsZentrum web site)). Also see:
https:en.wikipedia.orgwikiLaw of the iterated logarithm.
(4) One dimensional Random Walk https://mathworld.wolfram.com/RandomWalk1-
Dimensional.html .
2 The principle of Analogy has been used by Hilbert in a completely dierent context: When
he proved that Euclidean Geometry is consistent if Ordinary Arithmetic is consistent.
3I have called myself a Theoretical Physicicist only because I have a Ph.D in Theoretical
Physics, but this was long ago. I had majored in Mathematics as an undergraduate and
followed up a life-long study of mathematics both in work(research) and as a hobby.

Page 113
theorist starts from Peano's Axioms and the axioms of logic with these as a basis
the Geometrician/ Mathematican discovers new theorems which are essentially
logical deductions from the Axioms. But a Theoretical Physicist starts from
the study of a phenomena as observed in nature or as viewed in an experiment.
He then tries to formulate laws which can explain this phenomena. Of course,
there are limitations in both view points: The Mathematicain can never dis-
cover anything which lies beyond the reach of his axioms (this apect has been
spectaculary demonstrated by Godel by his Incompleteness Theorem) and the
Physicist can never discover any Law without having had the opportunity to
view the phenomena or conduct an experiment.

So let me rst summarize what has been done:

SUMMARY OF WHAT HAS BEEN DONE

1. It has been shown that for RH to be true, the condition given by Little-
woods theorem regarding the function L(N ) must hold.
In fact the real diculty is that we are given only One instance of λ(n)0 s.
So if you (temorarily) x N , then L(N ) is the distance traversed in an N step
random walk. But this sequence is un-alterable and xed whereas for an N
step random walk we can have many random walks each of N steps, i.e we
can have many instances. The real challenge is: We are given only one
instance of L(N) which is xed, 4 then how can we compare with coin
tosses when we are given only one sample? The trick is that we are given
0
only one instance of λ(n) s but this is a sequence innitely long! We fully
exploit this, because an innite string (sequence) has innite information: we
can sample dierent lengths of the string (sequence) at dierent locations and
compare with an equal length of coin tosses (e.g performing χ2 tting etc ).
It is thus, by exploiting the innite information available we could
unravel the conundrum that was the RH.
2. In Section 2 of Appendix VI of the Main paper, I have calculated consec-
utive λ(n)0 s forming a large sequence, denoted by Λ[N0 , M ], of length M of the
form Λ[N0 , M ] ≡ {λ(N0 ), λ(N0 + 1), λ(N0 + 2), ..., λ(N0 + M − 1)}.where N0 is
some large integer.
It has been shown by actually computation and performing a χ2 t using
5
the methods suggested by Donald Knuth, that very large sequences containing
consecutive values of λ(n)0 s very closely resemble and are infact statistically
indistinguishable from coin tosses of equal length. I have done very many
computations (using Mathematica) and some of them have been presented in
the Tables in Appendix VI. These are accurate and actual computations and
the results are indisputable.
p
3(a) The sets of consecutive
√ p λ0 s denoted as S+ (N ) = Λ(N + 1, N ) and
S− (N ) = Λ(N − N + 1, N ) , (N a square integer) have the property of
being statistically indistinguishable from coin tosses. The reason for this was
demonstrably argued because each integer n occuring in the argument of λ(n)
4 The λ−sequence was xed ever since the time God made the integers!
5 Knuth D.,(1968) `Art of Computer Programming', vol 2, Chap 3. Addison Wesley

Page 114
in one of the sets sayS+ (N ) belongs to a dierent Tower 6 (see Appendix VI).7
∞ 2 2
(b) In fact the union of all such sets i.e. ∪k=1 (S− (k ) ∪ S+ (k )) covers the
0
entire set of consecutive λ(n) s right upto innity.
4. It has been shown that even though the lambdas satisfy the determin-
istic condition λ(m.n) = λ(m).λ(n) the sequence of consecutive λ(n)0 s remain
statistically indistinguishable from coin tosses for large n.
5. It was shown that for large n the λ(n)0 s satisfy the three criterion of (i)
Equal probability (ii) unpredictability, (iii) independence and therefore using
this Principle of Statistical Analogy, one can deduce
√ L(N ) has the asymptotic
C. N behavior for large N, just like a random walk. Thus proving RH.
6. By using Khinchin and Kolmogrov's formulas of the iterated logarithm,
a more accurate bound for the width of the ctitical line has been be obtained.
7. Additionally, an independent proof for showing the square root behaviour
of L(N ) starting from Littlewood's Anzats was also obtained (Appendix V).
Some theorems had alternative proofs.
8. Finally, the methods used not only proved RH but also gave reasons
for the observed phenomena (from numerous numerical computations) of the
consecutive λ(n)0 s behaving like a sequence of coin tosses, viz para 2 and paras
3(a) and 3(b) above.

CONCLUSION
A perusal of the above Summary shows how the Riemann Hypothesis
is intimately connected with the factorization of integers as depicted in the
λ−sequence and in the observed behaviour of the randomness of nite sets
(e.g. S− (N ) , S+ (N )) of consecutive λ(n)0 s, as demonstrated by numerous
8
computations (Appendix VI) . Further, the connection with RH arises from
PN
Littlewood's Theorem, which relates the growth of L(N ) = n=1 λ(n) to the
position of the critical line. Hence by investigationg the randomness of the λ(n)0 s
and the randomness of coin tosses in particular by comparing the equations (1)
and (3) with (2) and (4) respectively, and by making a through study of their
statistical similarities and using the Principle of Statistical Analogy we nally
arrived at the TRUTH of RH.
Regards
K. Eswaran


6 The concept of Towers was introduced to understand, in an intutive way, the behavior
of L(N )as depicted in the Figure 1 (see the Main Paper)
7 These S+ (N ) and S+ (N ) can be very large, for example choosing N = 10200 the set
sets
S+ (10200 ) has a length of a googol ( 10100 ), i.e. it is a sequence containing googol consecutive
λ(n)0 s, guaranteed to be statistically indistinguishable from a googol number of consecutive
coin tosses!
8 These numerical computations and χ2 correlations are very real and are actually present
and give very strong indications of randomness present in the λ(n)0 s which were actually
computed by the factorization of integers n. I strongly believe this phenomena has to be
explained by the Pure Mathematicians and not brushed aside or put under the carpet or
carelessly labelled as mere coincidence! I believe my papers provides the raison d'etre for
the existence of this phenomena.

Page 115
 YOUR PREVIOUS EMAIL of March 30 2020

Dear Professor Eswaran,

In my previous messages I tried to show you that your arguments are not
very convincing. Your proof consists of the following steps:
1. You dene predictability of a sequence in the following way:
The sequence an is predictable if for some integer M the term an (for large
$n$) depends on $M$ precedent terms.
This denition coincides with the denition of a recurrent sequence, used in
several parts of mathematics.
2. You state that a sequence which is not predictable is independent, without
dening the word "independent".
3. You recall a theorem from probability theory describing asymptotical
behavior of the expected value of a series of independent random variables. I
am not quite sure whether you quoted it correctly. I now it only for the case
when the random variables attain the values 0 and 1. In the case when you have
1 and -1 it seems that the expected value would tend to zero. But I am not an
expert in probability.
4. You state that this theorem implies that if a sequence is unpredictable
(in the above sense) and attains the values 1 and -1 with the same frequency,
√
then the sum of its rst N values is asymptotic to c N with some c.
5. Applying this to the sequence λ(n) one obtains the Riemann Hypothesis,
as it is well-known that if for all large N one has

N
X √
λ(n) ≤ c N , (∗)
n=0

then RH follows.
You do did not a proof the step 4, and actually the assertion stated there is
incorrect, as one can easily construct a sequence an with an = 1or − 1, which is
unpredictable in the sense of 1, attains its values with the same frequency but
does not satisfy the assertion (*). Example:
Fork = 1, 2, . . . denote by Ik − 1) · 104 , k · 104 − 1] and denote
the interval[(k
3/4
byAk the set of all even integers in Ik to whichk · 103 odd integers from
Ik , taken at random, are added, and let Bk be the set of all remaining integers
from Ik . Let A be the union of all sets Ak and B the union of all sets Bk , and
dene A(x), B(x) to be the numbers o elements of the sets A, B in the interval
[1, x].
Dene a function f (n) by f (n) = 1 for n ∈ A and f (n) = −1 forn ∈ B ,
and observe that because of the randomly added elements to B the functionf
satises your unpredictability condition.
One sees immediately that we have

lim A(x)/x = lim B(x)/x = 1/2,

x→∞ x→∞

and

Page 116
 X
f (n) = A(k · 104 ) − B(k · 104 ) = k 3/4 · 103 (∗∗)
n≤k·104
√
but according to your assertion 4 this sum should be bounded by
√ c k · 104
= c k · 102 in contradiction to (**)
Therefore your proof is not correct.
Do not worry, your are not the rst person who presented an incorrect proof
of RH. At their list one nds some excellent mathematicians.
With every good wish
Wladyslaw Narkiewicz
END OF YOUR EMAIL 30th March 2020.


Page 117
 MY PREVIOUS EMAIL of March 17, 2020
Letter to Prof
Dear Professor Wladyislaw Narkiewicz, Wladyslaw
I now dene predictability of the λ0 s follows. Narkiewicz
DEFINITION: We say that λ(n) is predictable if, there exists a nite integer
M , such that for every n > N0 , the λ(n) , is derivable from its M previous values
λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ).
Note, the value of n is not explicitly known: only the values {λ(n − r), r =
1, 2, ..M } are known.
We then assume that if λ(n) is not predictable then it is independent.
Of course, if λ(n), is derivable as above, then there exists some function f
s.t :

λ(n) = f (λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ))

The justication for these denitions are given in Section 1 and Section 2.

SECTION 1:
Regarding your objection to the treatment of Eq(1) viz.

N
|L(N )| = Σn=1 λ(n) (5)

which by Littlewood's Theorem is supposed to satisfy the relationship:

|L(N ) ≤ C N a+ε (6)

for large N. In (2) I have used a in the exponent on the R.H.S. instead of
1/2, according to Littlewood if it so happens that a = 1/2. then RH is proved.

Comparing the above with the summation of a sequence coin tosses (or
random walk) depicted by:

N
|SN | = Σj=1 Xj (7)

where the random variable Xj takes on values +1 or −1, with equal prob-
ability and are independent of each other then, it is well known that for large
N:

2√
r
|SN | ≈ N (8)
π
Now, we have already shown that the λ(n) has the probability of being +1
or −1 for very large n. In addition I have proved (in Appendix IV) that for
large n, there is no way of knowing λ(n + 1) λ(n).9 This
if we happen to know
was sucient to prove independence for large n. (Note if we know λ(n) the
next predictable value is λ(2n) = −λ(n) , and 2n is very far away from from n

9 In fact, in Appendix IV, we have an alternative proof of 'independence' because we showed

that the sequence of λ0 s between λ(n + 1) and λ(2n) are independent of each other; this length
of this strip becomes innite as n tends to innity.

Page 118
 n.) Later, I used another criterion of independence surmising that if
for large
λ(n+1) is derivable (predictable) from its M previous values, then the λ(n)0 s are
predictable and hence not independent. But I showed that no xed nite M ex-
ists
10 and hence we can consider the λ0 s are independent for large n. (Therefore,
this second criterion is more general because the rst corresponds to M=1). We
show in the second Section 2 that this denition of predictability is the correct
one because by proving the λ0 s are un-predictable (by this denition) and hence
0
independent is able to explain the behavior ('phenomena') of the λ s over large
consecutive sections.
Hence we see that by using the same logic which derives (4) from (3) we
can derive (2) from (1) and we will have the position of the critical vertical line
at a = 1/2 thus proving R.H. However, in the paper I use a yet more rigorous
analysis (see last paragraph of page 14 of the main paper) which was done in a
study on independent random variables done by Khechin and Kolmogorov using
iterative logariths, their analysis gives the dependence of ε on N and how as N
tends to innity. From their result I have shown that the width of the critical
line tends to zero, proving that all the nontrivial poles of F(s) and hence the
zeros of ζ(s) lie on the critical line. QED

has

SECTION 2 : Phenomenological study of behavior of λ(n)0 s

In Section2 of Appendix VI of the Main paper, I have calculated consec-
utive λ(n)0 s forming a large sequences of length M of the form Λ[N0 , M ] ≡
{λ(N0 ), λ(N0 + 1), λ(N0 + 2), ..., λ(N0 + M − 1)}.where N0 is some large in-
2
teger. And for each such case I did a χ t which compare sit with a Bino-
mial sequence (coin tosses) of the same length M. as described by Knuth . In
each case the value of χ2 ≤ 4.0 showing that the λ0 s in the sequence are in-
distinguishable from coin tosses. In the Tables given in the Appendix VI , I
√
typically choose N0 = N M = N . E.G. the items in
(a perfect square and
row 2 Table 1.4 show that the sequence of length M = 100, 000 starting from
λ(10, 000, 000, 001),......,to λ(10, 000, 100, 000) has a χ2 value equal to 1.15,thus
it is statistically indistinguishable from a sequence of 100, 000 coin tosses or a
random walk of 100, 000steps. This phenomena' depicting the statistical be-
0
havior of the λ s happens for all the sequences of ALL the entries in the Tables
in Appendix VI. The reason for this phenomena is because each λ in the set
√
Λ(N, N ) is sampled
√ from a dierent 'Tower' and computing the summatory
function L(N, N ) is like randomly picking a number from the next Tower
which then may have value +1(H) or -1(T) with equal probability.
11
Since all the sequences Λ[N0 , M ] , of diering values of N0 and M , behave
like coin tosses regardless of their starting value N0 , we can say that the statis-

10 If no xed nite M exists (i.e. M grows large with n or is innite) it automatically

means that the λ(n + 1) is all the more unpredictable from λ(n) thus the λ0 s are practically
independent for large n.
11 It is possible to argue similarly, and demonstrate that the sequence Λ(N0 , M ), for large
N0 and M < 2N0 will also statistically behave like coin tosses because each λ in the sequence
will be drawn from a dierent Tower. and could be randomly +1(H) or -1(T).

Page 119
tical behavior does not depend on the starting value N0 ,further the statistics of
Λ[N0 + k, M ] (k being a small number) is the same as that of Λ[N0 , M ] we call
this property as Translational Invariance which is certainly true if we are talk-
ing of the N0th coin toss c(N0 )(≡ XN0 ); the coin tosses are all unpredictable and
therefore independent. Now the numerous computed χ2 values show that the
0
sequences of the λ s are statistically indistinguishable and also have the property
of being statistically Translational Invariant for their corresponding sequences
of coin tosses. Hence, to actually explain this phenomena of the statistical be-
havior we must show that the λ0 s are also statistically Translational Invariant,
'unpredictable' and therefore 'independent'. Hence we are by tour-de-force led
to the following denitions:
DEFINITION: We say that λ(n) is predictable if, there exists a nite integer
M , such that for every n > N0 , the λ(n) , is derivable from its M previous values
λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ).
Note, the value of n is not explicitly known only the values {λ(n − r), r =
1, 2, ..M } are known.
We then assume that if λ(n) is not predictable then it is independent.
Of course, if λ(n), is derivable as above then there exists some function f s.t
:

λ(n) = f (λ(n − 1), λ(n − 2), λ(n − 3), .....λ(n − M ))

In the Main Paper we proved that such a function cannot exist otherwise it
will make the λ-sequence cyclic, proving un-predictability and hence indepen-
dence of the λ0 s.
From the above analysis we can conclude that as N tends to innity the
statistical behavior of the λ0 s and the 'coin tosses' become very similar. Thus
just as Eq.(4) follows from Eq. (3) for coin tosses, we are forced to conclude
that Eq. (2) (with a=1/2) follows from Eq. (1) for the λ0 s .
12
We have thus shown that the λ-series has a dominant statistical behavior of
a random walk and therefore the summatory function L(N ) can be computed
as is done in the paper using the iterative logarithm formula derived by Khechin
and Kolmogorov, thus proving R.H.
It cannot be over emphasised that apart from proving RH, I have given
reasons for the phenomena of large subsets sets of consecutive
p √ p λ0 s denoted as
S+ (N ) = Λ(N + 1, N) and S− (N ) = Λ(N − N + 1, N ) , (N a square
integer) behaving like coin tosses, (see Eqs. (13) and (14), Sec. 4, Appendix VI
of Main paper). This is important because by taking a collection of all perfect

12 This λ(p.n) = λ(p).λ(n) = −λ(n), which can be used

is because the deterministic formula
λ√0 s having known λ(n) does not disturb the statistics of any
to predict the other values of the
large sequence, for example Λ[N, N ] (N being square integer), this is because even if λ(n)
belongs to the sequence, even the very next√predictable value λ(2n) lies outside the range
of the sequence and does not occur in Λ[N, N ]. The predictive power of the deterministic
formula for λ(m.n) = λ(m).λ(n) , causes only a minor perturbation on the dominant statistical
'random walk' behavior of L(N), for very large N, and whose eect fades away at innity.
In my project log in my Researchgate webpage, I have examined the behaviour of λ0 s and
its eect on L(N) and they have conrmed this assertion.

Page 120
squares N, the sets S+ (N )and S− (N ) contain all the λ0 s upto innity. Since,
the Tables given in Appendix VI are actual computed values of the λ0 s we have
not only proved RH but have given an explanation of the behaviour of the λ0 s .
Therefore, I believe that: Even if one is hard pressed to deny this proof of RH,
then he/she is compelled to deny the very existence of this phenomena, but the
latter cannot be denied because too many computations and χ2 comparisions
have been conrmed. It is because of these reasons (and of course my own
intution), I humbly believe that, what we have is a proof of RH!
In this reply I have conned myself to only your objections raised as points
1 and 2. I am leaving out point 3, as you have said a proof by von Mangoldt
already exists. (I have given (in Ref 3, in the 1st Invitation Letter (email) dated
3rd Feb 2020, an alternative proof which relies on induction and not mapping,
but I acknowledge that it has now become academic and redundant!).
To conclude, Professor Wladyslaw Narkiewicz, I want to once again express
my thanks to you for sparing so much of your time reading my work. I do
not know how to express my gratitude except by taking every word of yours
seriously (thus demonstrating my profound respect for you) and try to make a
complete clarication to the best of my ability. I will be most happy to clear
any further doubts (if any). God Bless!
Regards
Kumar Eswaran
March 17, 2020

END OF EMAIL of 17th March 2020

10

Page 121
 - ---------

"Władysław
Narkiewicz" <Wladyslaw.Narkiewicz@math.uni.wroc.
pl>
to: "Dr. Kumar Eswaran"
<kumar.e@gmail.com
>

date: Apr 3, 2020, 8:55 PM

subject: Re: Response to your
Email of the 30th
March Re: Riemann
Hypothesis
mailed-by: math.uni.wroc.pl

Dear Professor Eswaran,

Thank you for your quick response. This thime I send you a short message
with only one question. You find it in the attachment.

With best wishes

Wladyslaw Narkiewicz

---
SEE ATTACHMENT: From_Wladyslaw_April_3_2020_A-Short_Comment.pdf

------

Page 122
 p √
1. You assert that the theorem about random walks giving asymptotics 2/π N for the limit of
|E(SN )|/N (your Equation (4)) implies for every ε > 0 the inequality
N
X
λ(n) ≤ c(ε)N 1/2+ε
n=1

with some positive c(ε).

In your messages you gave also some vague indications of the proof, for example in your last message
you wrote:
”Therefore, any other sequence consisting of numbers +1 or −1 which satisfies the three above criteria
(i), (ii) and (iii), will have its sum of N terms behave like (4)”, adding that there is no need of ”an elaborate
machinery of probability spaces”.
Remember that there are two meanings of the word ”probability”. One, the common sense meaning,
is applicable to events occurring in real life, and the other, used in mathematics, lives only in probability
spaces. You must have a probability space Ω, a family of its subsets X ⊂ Ω and a function p(X) with
values in [0, 1] satisfying certain conditions, otherwise one is unable to apply probabilistic theorems in other
branches of mathematics. It seems that the problem of your proof lies in the fact that you disregard the
difference of these two notions of probability.
If you really have a proof of your assertion, then I would like to be able to see it.
2. You pointed out that my example is wrong, but you did not observe that in the set A containing
even numbers there are also bunches of odd numbers inserted at random, hence from f (n) = 1 the equality
f (n + 2) = 1 follows not always.

Page 123
Dr. Kumar
Eswaran <kumar.e@gmail.com
>
to: Władysław Narkiewicz
<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

bcc: "Dr. Kumar Eswaran"

<drkumar_eswaran@yahoo.com>

date: Apr 5, 2020, 12:48 PM

subject: Re: Response to your Email of April 3rd
Dear Prof Wladyslaw Narkiewicz,

I thank you for your email of the 3rd April..

I have answered to my best of my ability the queries that you have raised.
Please see attached.

Also attached is an Extract of the bench mark paper by Prof S. Chanrasekar's (Nobel
Laureate).I have made some annotations, for your convenience.

Thank You very much for your interest.

Regards
Kumar Eswaran

SEE ATTACHMENT: April_5_2020_Reply_Five_to_Wladyslaw.pdf

------------

Page 124
 MY WRITE UP IN RESPONSE TO YOUR EMAIL of 3rd April
2020
5th April 2020

Dear Professor Wladyslaw Narkiewicz,

I thank you for the trouble you are taking over my papers.

1(a) I rst address your rst query in your para 1:

In this section I provide a gist of the requested proof, most of the details are
in the Main paper and the Extended Abstract.
At the outset I wish to say that the summatory function L(N ) :

N
|L(N )| = | Σn=1 λ(n) | (1)

and the summation of a sequence coin tosses (or random walk) depicted
by:

N
|SN | = | Σj=1 Xj | (2)

(where the random variable Xj takes on values +1 or −1, with equal prob-
ability), will behave statistically identically for large N, if ALL the following
conditions are satised by λ(n) viz. (i) Equal probabilities, (ii) independence
(each toss is independent of the previous toss). The second condition eec-
tively means unpredictability' in Random Walk terms, (see Extended Abstract
If all these conditions are satised then only we can con-
for details).
clude that, for large N:

|L(N ) | ≤ C N 1/2 +ε (3)

Before answering your other queries, I wish to say that if there is a ran-
dom variable Xj which takes on values +1 or −1, with equal probability and
is independent i.e. they satisfy the criterion: (i) Equal probabilities and (ii)
independence then, it can be proved that, for large N,
√
|SN | ≈ C N (4)

The proof of Eq.(4), following from the denition Eq.(2) and the criteria (i)
and (ii) is a very well known result in the Statistical theory of random variables
pertaining to coin tosses or 1-d random walks. See references in Footnote:
1

1 See e.g. (1) Chandrasekhar S., (1943)'Stochastic Problems in Physics and Astron-
omy',Rev.of Modern Phys. vol 15, no 1, pp1-87
(2) Khinchine, A. (1924) \Uber einen Satz der Wahrscheinlichkeitsrechnung", Fundamenta
Mathematicae 6, pp. 9-20,
(3) Kolmogorov, A., (1929) \Uber das Gesetz des iterierten Logarithmus". Mathema-
tische Annalen, 101: 126-135. (At the DigitalisierungsZentrum web site)). Also see:
https:en.wikipedia.orgwikiLaw of the iterated logarithm. √
NOTE: Ref.(2) and Ref.(3) gives a more accurate version of the c. N behavior,
my Main Paper contains the details.
(4) One dimensional Random Walk https://mathworld.wolfram.com/RandomWalk1-
Dimensional.html .

Page 125
 I am enclosing as a separate Attachment to this Email, an extract of the
rst 3 pages of Prof S. Chandrasekar's bench-mark paper where he deals with a
1-dimension Random Walk, pl. see pages 2 and 3 of this attachment the section
The Simplest 1- dim. Random Walk Problem.
In his paper he shows for a Random walk: Eq.(4) follows from Eq.(2), notice
that he uses only the two criteria (i) Equal probability and (ii) Independence
to prove his result. I have Annotated and added a sticky notes on page 2 and
3, indicating the places where Chanrasekhar uses only the (i) and (ii) to derive
Eq.(4) from Eq.(2).
Now comparing Eq.(1) and Eq.(2) we have the challenge that Eq.(1) repre-
sents a single instance of one sequence of λ0 s which is given to us.
2 Where as
Eq.(2) represents a generic case of a random walk (or coin tosses) and there can
be many such random walks. The challenge is how, when we are given only one
instance, can we extract its statistics? We overcame this challenge by exploiting
the fact that though there is only one instance of Eq.(1) that is given to us, it is
a sequence of innite length. Therefore, we can examine very many sections of
it and also use the fact that the λ(n)0 s have been obtained by the factorization
of integers and are therfore governed by the laws of Arithmetic. By exploiting
this fact we were able to prove (in the Main Paper) that the λ−sequence has
(i) Equal probability and (ii) Independence.
But Chandrasekhar had shown that a sequence consisting of num-
bers +1 or -1, which satises the criteria (i) Equal probability and
(ii) Independence, will have its sum of N terms behave like (4), and
therefore (1) being an instance of a sequence which follows (i) and (ii)
will also behave like (3) (because (4) and (3) are actually mean the
same thing - diering only in notation). For convenience I call this as
the Principle of Statistical Analogy. 3 To repeat: Eq. (3) will follow from
Eq. (1), because we have already proved (in the Main Paper) that the λ(n)
satises the criteria (i) and (ii). Thus RH necessarily follows from Littlewood's
theorem. QED.
Additional Note: Littlewood's Theorem only requires that these criteria
need only be satised by the λ-sequence for large N and we have proved that
this is so.

1(b) Regarding your query on the denition of probability:

I dene probability in the same manner as Prof. Chandrasekhar has done.
Since a coin toss Xj can take values [H, T ]≡ [−1, 1] and this is similar to the
values that λ(n) can take viz. [−1, 1] there will be no contradiction.
All the denitions adopted by me has the virtue of being natural denitions
and are simple and sucient for my purposes: which was to show that the
λ-sequence satisfy the criterion (i) and (ii).

2 The λ−sequence was xed ever since the time God made the integers!
3 The principle of Analogy has been used by Hilbert in a completely dierent context: When
he proved that Euclidean Geometry is consistent if Ordinary Arithmetic is consistent.

Page 126
 2. Coming to your second query labeled 2:

Your sequence certainly does not satisfy independence, because most of the
time if n is odd f (n) = f (n+2) and all of the time when n is even f (n) = f (n+2).
The sequence will f ail the test of independence. I simply cannot imagine that
any person can experimentally toss coins for any length of time and get your
result. If you still remain unconvinced on what I say, I suggest you look up the
statistical tests of independence or ask a statistician.
The second fact that was overlooked (I forgot to mention this in my last
email) when the counter Example was formulated, is that one must remember
that the sequence {λ(n), n = 1, 2, 3, ...., ∞} is not any arbitrary sequence: It is a
sequence where for every integer n the value λ(n) depends upon the number of
prime factors of n and λ(n) = +1or −1 whether the number of prime factors are
even of odd (multiplicities). Therefore the actual string {λ(n)} is determined
by the rules of arithmetic used for factorization a number using primes. For
example: Consider the λ−sequence of 11 consecutive integers n (say) from 18
to 28,
{λ(18), λ(19), λ(20), λ(21), λ(22), λ(23), λ(24), λ(25), λ(26), λ(27), λ(28)}
={λ(2.3.3), λ(19), λ(2.2.5), λ(3.7), λ(2.11), λ(23), λ(2.2.2.3), λ(5.5), λ(2.13),
λ(3.3.3), λ(2.2.7)}
= {−1, −1, −1, +1, +1, −1, +1, +1, +1, −1, −1} .
Your sequence f (n) for (n even) satises f (n) = f (n + 2) but the condition
λ(n) = λ(n + 2) can never happen in other words the rules of arithmetic and
factorization cannot permit such a function f (n) for any reasonable stretch of
length, as proposed.
So one must remember that when one wishes to construct a valid counter
example, one must take care of (i) equal probabilities, (ii) independence (i.e no
correlation) and (iii) also remember that the λ−sequence is determined by rules
of factorization,
4 any arbitrary sequence simply will not do!


3. (a) Conclusion
Professor, out of regard to your high reputation of being a very good Number
Theorist and because of the interest and trouble you have taken in reading my
work I have considered every query of yours very seriously. I have as far as I can
tell, have answered all your queries and I have actually nothing more to add.
If, perchance, you have some diculties with the statistical and other methods
employed I suggest you could ask a statistician.

4 The rules of factorization govern the value of every single λ(n) but the rules of arithmetic
work in such a manner that the sequence {λ(n), n = 1, 2, 3...., ∞ becomes indistingusable from
coin tosses as n appraches ∞.

Page 127
 3(b) Regarding Query
Now because you are a number theorist I make myself bold to ask one Query.
Therefore, after perusing the following preamble I would be grateful if you could
please attempt to answer my Query at the end of it.
Preamble to Query:
In Section 2 of Appendix VI of the Main paper, I have calculated consecutive
λ(n)0 s forming a large sequence, denoted by Λ[N0 , M ], of length M of the form
Λ[N0 , M ] ≡ {λ(N0 ), λ(N0 + 1), λ(N0 + 2), ..., λ(N0 + M − 1)}.where N0 is some
large integer.
It has been shown by actually computation and performing a χ2 t using the
5
methods suggested by Donald Knuth, that these very large sequences contain-
ing consecutive values of λ(n)0 s very closely resemble and are in fact statisti-
cally indistinguishable from coin tosses of equal length. I have done very many
computations (using Mathematica) and some of them have been presented in
the Tables in Appendix VI,sec.3, e.g. Tables 1.3 and 1.4 page 27. These are
accurate and actual computations and the numerical results are indisputable .
6
By very many numerical computations I have shown that the sets of consec-
p √ p
utive λ0 s denoted as S+ (N ) = Λ(N +1, N ) and S− (N ) = Λ(N − N +1, N )
, (N a square integer) have the property of being statistically indistinguishable
from coin tosses.
7 The reason for this was demonstrably argued because each
integer n occurring in the argument of λ(n) in one of the sets say S+ (N ) belongs
to a dierent Tower
My Query:
These numerical computations and χ2 correlations are very real and are ac-
tually present and give very strong indications of randomness present in the
λ(n)0 s which were actually computed by the factorization of integers n. I
strongly believe this phenomena has to be explained by the Pure Mathemati-
cians and not brushed aside or put under the carpet or carelessly labeled as mere
coincidence!
Now I respecfully ask you, as a Number Theorist, how do you
explain this phenomena of the random behavior of the λ−sequence?
Regards
Kumar Eswaran

5 Knuth D.,(1968) `Art of Computer Programming', vol 2, Chap 3. Addison Wesley

6 I believe my papers provides the raison d'etre for the existence of this phenomena.
7 Notice that if you choose N = j 2 then the union of the two sets :
S+ (j 2 + 1, j) ∪ S− ((j + 1)2 − j, j + 1) is nothing but the sequence
{λ(j 2 +1), λ(j 2 +2), λ(j 2 +3), ..., λ((j +1)2 )}. That is they cover all the λ0 s with arguments
between two consecutive perfect squares, j 2 to (j + 1)2 . Now if you choose N = (j + 1)2 you
can cover the next region between the perfect squares ((j + 1)2 + 1) to (j + 2)2 and therefore
you can capture all the regions between two consecutive perfect squares all the way up to
innity - basically covering all integers by the union of sets S− and S+ right up to innity.

Page 128
"Władysław
Narkiewicz" <Wladyslaw.Narkiewicz@math.uni.wroc.pl
>
to: "Dr. Kumar Eswaran"
<kumar.e@gmail.com>

date: Apr 14, 2020, 11:21

PM
subject: Re: Response to your
Email of April 5
Dear Professor Eswaran,

I read very carefully your last message, but it did not convince me that
you really have a proof of Riemann Hypothesis, as your arguments are of
heuristical nature. I agree that the similarity of the considered sequence
of values of the lambda-function with a random walk gives some reasons
to believe in the truth of the conjecture. A similar idea appears already
in the literature. In the attachment you will find a paper which perhaps
will be of interest to you. It has been written by Good and Churchhouse
published in the journal Mathematics of Computation (vol. 22, 1968,
857--861) a time ago with a similar heuristical approach to the Riemann
Conjecture, based on the sequence of non-zero values of the Moebius
function.

With best wishes

Wladyslaw Narkiewicz
SEE ATTACHMENT: Paper_sent_by_Wladyslaw_GoodChurchHouse.pdf

Page 129
The Riemann Hypothesis and Pseudorandom Features of the Möbius Sequence
Author(s): I. J. Good and R. F. Churchhouse
Source: Mathematics of Computation, Vol. 22, No. 104 (Oct., 1968), pp. 857-861
Published by: American Mathematical Society
Stable URL: https://www.jstor.org/stable/2004584
Accessed: 14-04-2020 17:47 UTC

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2004584?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend
access to Mathematics of Computation

This content downloaded from 156.17.86.7 on Tue, 14 Apr 2020 17:47:22 UTC Page 130
All use subject to https://about.jstor.org/terms
 The Riemann Hypothesis and
Pseudorandom Features of the Mobius Sequence

By I. J. Good* and R. F. Churchhouse

Abstract. A study of the cumulative sums of the 1\Tbius function on the Atlas
Computer of the Science Research Council has revealed certain 'statistical prop-
erties which lead the authors to make a number of conjectures. One of these is
that any conjecture of the Mertens type, viz.
N

IM(N) I = E 4(n) < k(-\IN)

n=1

where k is any positive constant, is false, and indeed the authors conjecture that

Lim sup {M(x)(x log log x)-112} = V (12)/r .

The Riemann zeta function is defined for R(s) > 1 by the series c(s) =
E=li nB-, and the definition is completed by analytic continuation. Rie
hypothesis is that the "complex zeros" all occur where R(s) = 2. It has now been
verified for the first 2,000,000 complex zeros (Rosser and Schoenfeld, [8]), but this
is not a very good reason for believing that the hypothesis is true. For in the theory
of the zeta function, and in the closely allied theory of the distribution of prime
numbers, the iterated logarithm log log x is often involved in asymptotic formulae,.
and this function increases extremely slowly. The first zero off the line R(s) = 2
if there is one, might have an imaginary part whose iterated logarithm is, say as
large as 10, and, if so, it might never be practicable to find this zero by calculation.
The plausibility of this argument is increased when we recall the refutation by
Littlewood [5] of the conjecture that 7r(x), the number of primes less than x, is,
always less than the logarithmic integral, li(x), a conjecture that is presumably-
true at least as far as x = 109 (see Ingham [4, p. 7]). It is possible that Littlewood
had this kind of argument in mind when he said (Littlewood [6]), "In the spirit of
this anthology (an anthology of partly baked ideas) I should also record my feel-
ing that there is no imaginable reason why it (the Riemann hypothesis) should be
true."
The aim of the present note is to suggest a "reason" for believing Riemann's.
hypothesis.
The M6bius function is defined by ,u(n) = (_)k if the positive integer n is the
product of k different primes, ,u(1) = 1, and ,u(n) = 0 if n has any repeated factor.
It is known (see, for example, Titchmarsh [9, p. 315]) that a necessary and suf--
ficient condition for the truth of the Riemann hypothesis is that M(x) - O(x' 12+e),*
for all e > 0, where M(x) = ,u(n) (n ? x). The condition M(x) - O(x1 2+E)
would be true if the M6bius sequence {,u(n) } were a random sequence, taking the
values -1, 0, and 1, with specified probabilities, those of -1 and 1 being equal.-

Received May 1, 1967.

* Now at Virginia Polytechnic Institute, Blacksburg.

857

This content downloaded from 156.17.86.7 on Tue, 14 Apr 2020 17:47:22 UTC Page 131
All use subject to https://about.jstor.org/terms
858 I. J. GOOD AND R. F. CHURCHHOUSE

More generally, if we first select a sub

the terms for which ,u(n) = 0, and if this subsequence were 'equiprobably random,'
i.e. if the value -1 and 1 each had (conditional) probability -1, then the condition
M(x) = O(x'12+e) would still be true. Of course a deterministic sequence can at
best be 'pseudorandom' in the usual incompletely defined sense in which the term
is used, and of course all our probability arguments are put forward in a purely
heuristic spirit without any claim that they are mathematical proofs.
As a matter of fact there are a priori reasons, that is without looking at the
numerical support, for believing the following conjecture:
Conjecture A. The sums of ,u(n) in blocks of length N, where N is large, have
asymptotically a normal distribution with mean zero and variance 6N/72.
A priori reasons for believing this conjecture. Note first that ,u(n) = 0 when n is
a multiple of 4, so there are zeros in the sequence {,u(n) } at regular intervals of 4,
and similarly at regular intervals of 9 and so on. So, if we write vP for the number
of values of n (in a block of length N) for which ,u(n) i, we would expect vo to
be very close to its expected value

N - N(1 - 2-2)(1 - 3-2)(1 5-2) ... = N(1 - 6r-2) .

The numerical evidence for this statement is given in Tables 2 and 3 in the appendix.
Now if n is large and known to be square-free it is likely to have a fair number
of factors, and therefore by the theory of the roulette wheel (with two sectors in-
stead of 37 or 38) the probabilities that the number of factors is odd or even are
nearly equal. Thus the probability that ,u(n) = 1 (or -1) is near to 37r-2 and tends
to this value when the range in which n is known to lie tends to infinity. Hence
the expectation of ,u(n) is 0.
The probability distribution of vj, conditional on a knowledge of vo, is binomial
with mean 2 (N - vo) and variance 4 (N - vo). Allowing for the near-constancy of
vo, the unconditional distribution of vi would be expected to be binomial with mean
3N7r-2 and variance 3N7r-2/2. As a matter of fact the variance does not depend on
the near-constancy of vo since

var (v1 - v-i) = E(V1-V_1)2

= E {E[(vl - v_1)2Ivo] }
= E(N - vo) = 6N7r-2.

We have here assumed that, given vo, v1 has a 'heads-and-tails' binomial distribu-
tion. Its sample size is, of course, N - vo.
This completes our a priori argument for believing Conjecture A, and even if
Conjecture A is only approximately true it is so much stronger than the condition
M(x) = O(xI'2+e) that we feel its approximate truth would still support that
condition.
We now describe the numerical test of Conjecture A, which was performed
with N = 1000.
We computed M(lOOOr + 1000) - M(lOOOr) (where we write M(O) = 0) for
r = 0(1)49,999 on the Chilton Atlas (as a "background" job) but the values
r = 34,000(1)34,999 were lost owing to a machine fault. Column (ii) of Table 1
gives the frequencies with which

This content downloaded from 156.17.86.7 on Tue, 14 Apr 2020 17:47:22 UTC Page 132
All use subject to https://about.jstor.org/terms
 THE RIEMANN HYPOTHESIS AND THE MOBIUS SEQUENCE 859

M(lOOOr + 1000) - M(lOOOr)

lies in the ranges given by column (i). The "expected values" of these frequencies,
based on Conjecture A, are listed in column (iii). For example,

49 fl0.6
7688 = V2 exp (- '2X2/C
where c2 = 6000/7r2. It will be seen that the fit is extremely good, in fact x2 = 24.2
with 22 degrees of freedom. Thus Conjecture A is not merely to be expected a
priori, by mathematical common sense, but it is well supported by the numerical
data. As we said before, we believe therefore that there is a good "reason" for
believing the Riemann hypothesis, apart from the calculation of the first 2,000,000
zeros.

TABLE 1.

Frequencies of the values of ,u(n) (10OOr < n ? 1000(r + 1)),

for 49000 values of r.

(i (ii) (iii)
Ra,n,,qe Frequencies Expectations

-lOlto -110 0 1.0

- 91 to -100 1 4.9
-81 to - 90 16 20.6
- 71to- 80 63 76.5
-61 to - 70 248 240
- 1 to- 60 635 652-
-41 to - 50 1491 1465
- 31 to - 40 2797 2832
- 21 to - 30 4711 4645
-11 to - 20 6604 6478
- 1 to - 10 7513 7688
0 793 794
1 to 10 7779 7688
11 to 20 6450 6478
21 to 30 4601 4645
31 to 40 2842 2832
41 to 50 1477 1465
51 to 60 651 652
61 to 70 216 240
71to 80 88 76.5
81 to 90 16 20.6
91to 100 7 4.9
101 to 110 1 1.0
49000 49000.0

Although {,u(n) } is not a random sequence it is tempting to apply the law of

the iterated logarithm (for example, Feller [1, p. 157]) to the subsequence obtainedt
by deleting the values of n for which i(n) = 0. In this manner we generate a.
second conjecture, which, however, is less probable than Conjecture A and for
which it is difficult to obtain numerical support. But it is of some interest to con-
sider it.

This content downloaded from 156.17.86.7 on Tue, 14 Apr 2020 17:47:22 UTC Page 133
All use subject to https://about.jstor.org/terms
860 I. J. GOOD AND R. F. CHURCHHOUSE

Conjecture B.

limsup {M(x)(xloglogx)- /2J = V (12)/7r.

Conjecture B contradicts Mertens's conjecture that IM(x) I < x"12, even in the
extended form IM(x) I< Cx"12 for any constant C. When C = 1 this modification
of Mertens's conjecture was refuted numerically by Neubauer [7]: a breakdown
occurred, for example at x = 7.76 X 109. Also a conjecture of P6lya's, closely
related to that of Mertens, was refuted by Haselgrove [3], who believed further
that his method could be applied, with 1000 times as much calculation, to disprove
Mertens's conjecture. In the light of this evidence, Mertens's conjecture is im-
probable, and our Conjecture B is somewhat supported by its inconsistency with it.

Appendix. Distribution of Po for N = 1,000,000

In Table 2 we give the values of Po, i.e. the number of cases of ,u(n) = 0, in the
first, second, . . ., 33rd block of length a million. We stopped at this point owing
to the machine fault previously mentioned: a single supervisor fault caused the
output for Table 1 to be lost at the 35th million and for Table 2 at the 34th million.

TABLE 2

Million PO Million Po

1 392,074 18 392,088
2 392,049 19 392,039
3 392,104 20 392,037
4 392,037 21 392,072
5 392,103 22 392,084
6 392,076 23 392,096
7 392,053 24 392,047
8 392,101 25 392,096
9 392,061 26 392,071
10 392,051 27 392,071
11 392,073 28 392,065
12 392,078 29 392,079
13 392,073 30 392,065
14 392,095 31 392,083
15 392,083 32 392,077
16 392,093 33 392,076
17 392,057

Using the 33 values of Po given in Table 2, the estimated standard deviation is

only 19.1. (The average number of zeros in each block of a million is very nearly
392,073.) This suggests that nearly always Po is nearly constant in the sense that,
for large N, the standard deviation of Po is o( -V N), which for possible future ref-
erence we call Conjecture C. The total expected number of zeros of the sequence
{,i(n)} in the first 33,000,000 is 33,000,000(1 - 6r-2) = 12,938,405.6 and the
observed number is 12,938,407, an astonishingly close fit, better than we deserved.
The values of Po for a few further values of N are shown in Table 3.
On the basis of this table we might even strengthen Conjecture C to (Conjecture
D) 'the variance of Po for large N is a constant.'

This content downloaded from 156.17.86.7 on Tue, 14 Apr 2020 17:47:22 UTC Page 134
All use subject to https://about.jstor.org/terms
 THE RIEMANN HYPOTHESIS AND THE MOBIUS SEQUENCE 861

TABLE 3

N PO E(Vo) - E(0o)

25,000,000 9,801,820 9,801,822.5 - 2.5

50,000,000 19,603,656 19,603,645 +11
75,000,000 29,405,440 29,405,467 -27
100,000,000 39,207,306 39,207,290 +16

The conjecture of the near constancy of Po in each block of length N must be

interpreted in an average ('probabilistic') sense, whether or not it is expressed in
the form of Conjecture C. It would not be correct to assume that Po is always close
to its expected value; in fact it will sometimes though very rarely happen that
vO = N. This will happen, for example, in the block (M + 1, M + 2, *. ., M + N)
if simultaneously M-- -1 (mod 4), M- -2 (mod 9), M -3 (mod 25),
**,M _-N (mod PN2), where PN is the Nth prime. These congruences can be
solved by Sun-Tsu's theorem (see, for example, Good [2, p. 759]). The value of
M so obtained will be something like N2N. Thus the M6bius sequence {,u(n) } con-
tains arbitrarily long runs of zeros, but these long runs presumably occur extremely
rarely.

Atlas Computer Laboratory

Chilton, Didcot, Berkshire
England

Trinity College
Oxford, England

1. W. K. FELLER, An Introduction to Probability Theory and its Applications, Vol. 1, Wiley,

New York, 1950. MR 12, 424.
2. I. J. GOOD, "Random motion on a finite Abelian group," Proc. Cambridge Philos. Soc.,
v. 47, 1951, pp. 756-762. MR 13, 363.
3. C. B. HASELGROVE, "A disproof of a conjecture of P6lya," Mathematika, v. 5, 1958, pp.
141-145. MR 21 43391.
4. A. E. INGHAM, The Distribution of Prime Numbers, Cambridge Univ. Press, New York,
1932.
5. J. E. LITTLEWOOD, "Sur la distribution des nombres premiers," C. R. Acad. Sci. Paris,
v. 158, 1914, pp. 1869-1872.
6. J. E. LITTLEWOOD, "The Riemann hypothesis" in The Scientist Speculates, edited by Good,
Mayne & Maynard Smith, London and New York, 1962, pp. 390-391.
7. G. NEUBAUER, "Eine empirische Untersuchung zur Mertensschen Funktion," Numer.
Math., v. 5, 1963, pp. 1-13. MR 27 45721.
8. J. B. ROSSER & L. SCHOENFELD, "The first two million zeros of the Riemann zeta-func-
tion are on the critical line," Abstracts for the Conference of Mathematicians, Moscow, 1966, 8.
(Unpublished.)
9. E. C. TITCHMARSH, The Theory of the Riemann Zeta-Function, Clarendon Press, Oxford,
1951. MR 13, 741.

This content downloaded from 156.17.86.7 on Tue, 14 Apr 2020 17:47:22 UTC Page 135
All use subject to https://about.jstor.org/terms
Dr. Kumar
Eswaran <kumar.e@gmail.com
>
to: Władysław Narkiewicz
<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

date: Apr 16, 2020, 3:19 PM

subject: My Response to your Email of April of the
14th April
mailed-by: gmail.com
Dear Professor Wladyslaw Narkiewicz,

I thank you for email of the 14th April.

I have responded to it in the attachment, please see below.

I thank for initiating an illuminating discussion.

Keep Safe and God Bless

Regards
Kumar Eswaran

SEE ATTACHMENT: April-16_2020_Reply_Six_to_Wladyslaw.pdf

- ------

Page 136
 MY RESPONSE TO YOUR EMAIL of 14th April 2020

Dear Professor Wªadysªaw Narkiewicz,

I thank you very much for your email of the 14th April.
I have to state as follows:
1. I was aware of Good and Churchhouse of 1968. They have done some
numerical experiments supporting the speculation made by Denjoy in 1931 on
the Mobius series, I have referred to Denjoy in my Main Paper see References,
(Good and Church House have not acknowledged Denjoy).
Anyway, Good and Churchhouse are wrong for the following reasons:
(a) The Mobius series {µ(n), n = 1, 2.3...∞} cannot be directly compared
with a sequence of coin tosses because µ(n) can take one of three values namely
(1, −1, 0) depending on whether n has resp. even number of distinct prime
factors or odd number of distinct prime factors or has a prime-square factor
i.e. divisible by p2 where p is a prime. Whereas a coin toss can take only two
values viz: +1 (H) or -1 (T) Because of this, clear bounds cannot be obtained
N
to estimate the behaviour of the sum Σn=1 µ(n) for large N.
(b) More importantly, Good and Churchhouse have not even tried to
prove by using, the laws of arithmetic (or number theory) (i) Equal probability
and (ii) Independence.
Because of the above two counts: (a) and (b), their work does not count as
a proof of RH. To be fair to them, they themselves acknowledge this, and do
not claim that their paper has a mathematical proof.
However the same cannot be said of my work!
Reasons:
2. It has been shown by Khinchine & Kolmogorov that any sequence {Xj , j =
1, 2, ..} made up of random variables Xj , which has the two properties (i) Equal
probabilities of Xj = +1or −1 and (ii) Independence will satisfy the relationship
(for very largeN ):

N
| Σj=1 Xj | ≤ C N 1/2 +ε (A)
In my last Email I had enclosed, for your perusal, a Review paper by S.
Chandrasekhar (Nobel Laureate), which derives the above from rst principles
starting from (i) and (ii).
3. I have proved that the sequence {λ(n), n = 1, 2, ...∞} made by terms be-
longing to the Liouville function λ(n) which takes values +1 or -1, has precisely
these two properties viz:(i) Equal probabilities of λ(n) = +1 or − 1 and (ii)
Independence (for large N ): And therefore by the same logical argument must
satisfy a similar relationship (for large N):

N
| Σj=1 λ(j) | ≤ C N 1/2 +ε (B)

The proof took most of my paper and involved the proof of may theorems
for many of which I have given alternative proofs.
4. The way Eq. (A) follows from the properties (i) and (ii) for the sequence {Xj }
of +1's and -1's, implies that any other sequence {λ(n)} of of +1's and -1's

Page 137
which has been proved to satisfy the properties (i) and (ii) (for large N) will
necessarily obey Eq.(B) and thus RH follows. 1
This is a deduction made by using the principles of mathematical Logic.
(Simply put: It means two mathematical entities which have similar properties
will obey similar relationships). Therefore, with due respect, I wish to say that
denying the logic, putting every thing done in the paper under the carpet i.e
all the proofs, arguments and experimental evidence, and then dismissing them
by using a single epithet: heuristic, is equivalent to saying that mathematical
logic is an unreliable tool and cannot be trusted if implemented in mathematical
proofs. I simply cannot accept this stand because it completely destroys the very
foundations upon which the edice of mathematics and science are built.

In this email (and all my others) I have tried to expain that for my proofs, I
had used only number theory, complex function theory and statistical concepts
and mathematical logic throughout the length of my paper. And therefore I do
not consider them Heuristic at all! But if you still think so, there is nothing
more to be said.
With this I end.
I just want to thank you for initiating our illuminating discussion THANK
YOU!
I hope the Covid19 pandemic is not aecting you and you are safe. From
the news I found that nearly 8000 peope have been eected in Poland. Here in
India it is around 12,000 but the worrying factor is that it is doubling every 4-5
days. I live in Hyderabad. DO TAKE CARE! Regards
Kumar Eswaran 16th April 2020

P.S. I have read many papers on RH, but very few of them have taken into
account the randomness that appears in prime factorization. (The late Freeman
Dyson and others rst showed the connection of randomness of primes with the
zeros of the zeta function). Most of such papers were involved in numerical com-
putations and they did not investigate the reason for such random behaviour.
Firstly, why should all the zeros be on a (critical) line? Secondly, when the few
discovered the seeming evidence of randomness, they did not attempt, using the
principles of arithmetic and number theory, to nd the underlying reasons for
such random behaviour.
2 It is primarily because of this, all of them have failed 3
to prove RH. I was driven to nd the underlying reason for the appearence of
randomness. I have been luck to have stumbled upon a solution, which truth to
say, started o serendipitously.
1 Note: From Littlewood's Theorem, Eq.(B) must hold for RH to be True. In fact, if one
can invent a sequence which has the properties (i) and (ii) and possesss the properties of the
sequence {λ(n)}, but does not satisfy Eq.(B), then one would have actually disproved the
Riemann Hypothesis!
2 Though some of them e.g. Polya, thought that the imaginary part of the zeros could be
the eigenvalues of a Hermitian matrix and were driven to search for a quantum Hamiltonian
whose eigenvalues will deliver the zeros, however such attempts were fruitless.
3 In this particular case, I have found that reading old papers on RH, takes you far away
from the ideas that actually would have lead you to a solution. It is better to read the present
paper from a fresh stand point and not see through it as through the foggy windows of the
past!

Page 138
Władysław
Narkiewicz" <Wladyslaw.Narkiewicz@math.uni.wroc.pl
>

to: "Dr. Kumar Eswaran"

<kumar.e@gmail.com>

date: Apr 17, 2020, 12:49

PM
subject: Re: My Response to
your Email of April of
the 14th April
mailed-by: math.uni.wroc.pl
Dear Professor Eswaran,

Although our views on your result are somewhat different I want you to
know that I enjoyed our discussion which showed that the notion of a proof
may have different interpretations. I want also to stress that the word
"heuristic" has no negative meaning. A lot of work of really great
mathematicians has been performed in a heuristic way. This applies not
only to old times (Euler, Laplace, the Bernoullis, ...) but also to recent
times.

Because of the pandemic I sit at home since three weeks, but this permits
me to read all the books which I bought a time ago and did not have time
to read them earlier. I am trying also to do some mathematics but my age
limits my possibilities.

With every good wish

Wladyslaw Narkiewicz

---- - - - - - - - -
Dr. Kumar
Eswaran <kumar.e@gmail.com
>
to: Władysław Narkiewicz
<Wladyslaw.Narkiewicz@math.uni.wroc.pl>

date: Apr 18, 2020, 12:10 PM

subject: Thank You For your clarification and some
thoughts
mailed-by: gmail.com

Page 139
18/4/2020
Dear Professor Wladyslaw Narkiewicz,
Thank you for your prompt email and gracious clarifications. Your kind words have greatly relieved me!
Yes, when I was working on the RH and beginning to realize how important the study of the growth and
behavior of the lambda- sequence viz. {lambda(n), n=1,2,3…..N}, is to the problem of RH, I figured the
following:
(i) That some better strategy, than merely using Peano’s Axioms Dedekind-Peano's
Axiom's ,
was needed for the proof of RH, because
(ii) By solely using the Axioms to study the growth of the lambda-sequence up to
infinity ,one would need knowledge of ALL the primes. (The axioms can only be used, in this
case, to find primes and factors of each integer step by step).
All I can say is that I have been lucky in pursuing an alternative direction.
Your participation in our discussion has been very illuminating. Thank You.

And DO Take Care! And God Bless!

Regards
Kumar Eswaran.
----

Page 140
 Report on prof. K. Eswaran’s proof of the Riemann hypotesis
Germán Sierra
Instituto de Fı́sica Teórica UAM-CSIC, Madrid, Spain.
(Dated: 9 September 2020)

I. THE MAIN ARGUMENT

Professor Eswaran has proposed in references1,2 a proof of the Riemann Hypothesis (RH) that will be summarized
below. We first need some basic definitions. Let λ(n) be the Liouville function that is a completely arithmetic
multiplicative function equal to 1 (resp. −1) if the integer n is the product of an even (resp. odd) number of primes.
Let L(x) be the summatory function
X
L(x) = λ(n) . (1)
n≤x

The RH is equivalent to the following statement

1
L(x) = O(x 2 + ), ∀ > 0 , (2)

or alternatively

L(x)
lim 1 = 0, ∀ > 0 . (3)
x→∞ x 2 +
This result is due to Landau and also appears in Littlewood works.

Prof. Eswaran’s logic to prove the RH is based on demostrating eq.(2). The key argument consists in showing that
the function λ(n) behaves like a random walk, or a coin tossing game, in which case the sum of the outcomes up to
N , would behave asymptotically as L(N ) ' CN 1/2 . If this were the case then certainly the RH would be proven !!
A general remark concerning Eswaran’s approach is that the apparent random walk nature of the Liouville function
is a well known fact in analytic number theory (see for example3,4 ). It is heuristic and, according to most expectations,
unlikely to be true. The reason being that the prime numbers are deterministic objects, not random, as well as the
Liouville and other arithmetic functions. These objects seem to behave randomly but they do not. Their apparent
random nature has been very useful in the past to propose ”conjectures” but it is impossible to use them to ”prove”
anything about the primes or related quantities.
An example of the heuristic use of the prime numbers is the twin prime conjecture of Hardy-Littlewood. The
prime number theorem states that the number of primes below x behaves asymptotically as x/ ln x. The prime
numbers look random with a probability density given by 1/ ln x. However there are correlations between them that
are described by the Hardy-Littlewood conjecture according to which the number of prime pairs (p, p + 2), below x
behaves asymptotically as C2 x/ ln2 (x) with C2 a constant. This conjecture is based on heuristic arguments but there
is not a proof of it. There are many more examples of this type, which suggest that a proof of the RH based on
random models is very unlikely. Similar methods have been applied to the Riemann zeros which seem to behave as
the eigenvalues of hermitean random matrices. This idea has been used heuristically to conjecture the moments of
the Riemann zeta function, but again there are not rigorous proofs except for a few cases.
These are general considerations to suspect why a proof of the RH based on random models is likely to be false,
as has occurred in the past, but it does not exclude it. Below I present some arguments to show that, unfortutanely,
Eswaran’s approach fails to prove the RH.
The purported proof of the RH is based on the statement that the infinite set Λ ≡ {λ(1), λ(2), λ(3), . . . } is an
instance of an infinite random walk R = {1, −1, −1, . . . , } where every term behaves as a random variable xi = ±1.
For this to be the case the following conditions have to be satisfied (i) equal probabilities of λ(n) = 1 or λ(n) = −1 ,
(ii) the λ-sequence has no cycle and (iii) unpredictability. Condition (i) follows from the Theorem in1
Theorem 3: In the set of all positive integers, for every integer which has an even number of primes in its
factorization there is another unique integer, (its twin), which has an odd number of primes in its factorization.

Page 141
 2

This theorem is derived through an interesting partition of the positive integers into infinite sets Pm,p,u , called towers,
whose elements, when ordered increasingly, have λ-values that alternate between +1 and −1. This representation is
in turn used to provide a nice geometric interpretation of L(N ) as waves
P depicted in fig.1 of1 . These partition of the
s
natural numbers is also used to rewrite the Dirichlet series F (s) = n λ(n)/n as a sum over the partitions Pm,p,u .
But the main application of the towers is to provide a proof of theorem 3. Let me try to prove this theorem in a way
that is inspired in these towers but is more simple and that will be used later on.
Theorem 3 is equivalent to the statement that all the positive integers, Z+ , is the disjoint union of two sets Z+ + and
−
Z+ , defined as

Z+
+ = {n ∈ Z+ |λ(n) = +1}, Z−
+ = {n ∈ Z+ |λ(n) = −1} (4)

that have the same cardinal. This means that there is a one-to-one correspondence between these two infinite sets.
To proof this let us consider the pair of integers

n0 = 3e3 5e5 . . . peLL , e3 , e5 . . . , eL ≥ 0 (5)

n1 = 2 3e3 5e5 . . . peLL

that satisfy λ(n1 ) = −λ(n2 ). The values of ea , determine to which set, Z±

+ , the integers n0 and n1 belong to, but they
are certainly different. The couple (n0 , n1 ) form a twin pair, using the terminology of reference1 . We can progress
considering the integers

n2 = 22 3e3 5e5 . . . peLL , e3 , e5 . . . , eL ≥ 0 (6)

n3 = 2 3 53 e3 e5
. . . peLL

that, by the same arguments as above, form a twin pair. The process can be interated at infinity and then all the
−
integers can be organized in twin pairs such that finally Z+ = Z+ + ∪ Z+ . The number 2 seems to play an special role
in this construction but it does not. One could choose say 3, or any other prime number, and repeat the process
−
obtaining the same result, namely Z+ = Z+ + ∪ Z+ . Depending of the prime choosen one will obtain a different pairing
which would amount to a permutation of the one-to-one maps between two infinite sets with the same cardinal.
Following reference1 , a conclusion of Theorem 3 is
Theorem 3B : If n is an arbitrary positive integer,

1
Prob[λ(n) = +1] = Prob[λ(n) = −1] = (7)
2

Strictely speaking this is not a theorem but a loose way to express theorem 3. One reason is that the defini-
tion of probability used in (7) is not given explicitly. If the sets Z± + where finite dimensional then the one-to-one
correspondence between them would perhaps be expressed as in equation (7). But these sets are infinite dimensional.
A proper definition of probability, regarding the parity of λ(n), can be given by considering the first N integers and
counting the number of positive integers with the same value of λ(n),

N+ = #{n ≤ N |λ(n) = +1}, N− = #{n ≤ N |λ(n) = −1}, N = N+ + N− . (8)

The probabilities of finding λ = ±1 in the first N integers are given by

N+ N−
P+ (N ) = , P− (N ) = (9)
N N
whose difference is given by the summatory Liouville function L(N ),
P
N+ − N− n≤N λ(n) Λ(N )
P+ (N ) − P− (N ) = = = . (10)
N N N
The ratio in the last term of the r.h.s in the limit N → ∞ was proven by Landau to vanish (doctoral thesis in 18993 )

Λ(N )
lim =0. (11)
N →∞ N

Page 142
 3

Landau showed that this result is equivalent to the Prime Number Theorem (PNT). In terms of the probabilities (9)
we can write
1
lim P+ (N ) = lim P− (N ) = . (12)
N →∞ N →∞ 2
and interpret this result saying that an integer has equal probability of having an odd number or an even number of
distinct prime factors (see page 8 of reference3 ). The latter statement is formulated in (7) as a theorem, but it is not.
The theorem 3B is derived in1 from theorem 3 and the properties of the towers, but the derivation is not convincing.
It is true that within each tower the values of λ(n) alternate, but going up in the list of integers, one jumps from one
tower to another with no obvious pattern. It would be very interesting to use the tower construction to prove, not
Theorem 3B that as explained earlier is not well formulated, but the PNT, namely equation (12). This is a much
more easier target than the RH, but if achieved would represent an interesting result by itself.
The study of the properties of the Liouville function has been the focus on many works in the past that led to the
development of new mathematical techniques. An interesting reference regarding this topic is5 that among another
results, proposes a probabilistic interpretation of the function e−x/x L(x) that nevertheless is quite different from a
random walk.

II. CONCLUSION

In my opinion prof. Eswaran’s works do not provide a proof of the RH. The methods employed to tackle this difficult
problem do not even provide an alternative proof of the Prime Number Theorem. The idea of interpreting the Liouville
sequence as random variables is attractive, although not original, but unfortunately has not been substantiated. The
mystery of the primes remains untouched.

1
K. Eswaran, ”The Final and Exhaustive Proof of the Riemann Hypothesis from First Principles”, Research Gate, Vol. May
2018, https://www.researchgate.net/publication/325035649.
2
K. Eswaran, ”The Pathway to the Riemann Hypothesis”, Preprint March 2019. DOI: 10.13140/RG.2.2.17950.59208.
https://www.researchgate.net/publication/331889126.
3
P. Borwein, S. Choi, B. Rooney and A. Weirathmueller (eds). ”The Riemann Hypothesis: A Resource for the Afficionado
and Virtuoso Alike”, CMS Books in Mathematics, Springer (2008).
4
I. Good and R. Churchhouse, ”The Riemann Hypothesis and Pseudorandom Features of the Möbius Sequence”, Mathematics
of Computation. 22, 857 - 864 (1968).
5
Peter Humphries, ”The distribution of weighted sums of the Liouville function and Pólya’s conjecture”, Journal of Number
Theory 133 (2013) 545-582.

Page 143
Your Open Review of my paper on RH
from: Dr. Kumar
Eswaran <kumar.e@gmail.com>
to: German Sierra
<german.sierra@uam.es>
bcc: "P. Narasimha Reddy"
<nrriemann@sreenidhi.edu.in>
date: Oct 6, 2020, 9:55 PM
subject: Your Open Review of my paper
on RH
mailed- gmail.com
by:
Dear Professor German Sierra,

I thank you for the trouble you took in participating in the Open Review of my paper on RH.

I am attaching my reply to your comments.

Kindly read the document, I will be happy to receive your response.
Since I have referred to the Lectures given by Vinayak (on my proof) in my reply, I have
attached his lectures for your convenience.

Kind Regards
Kumar Eswaran
Professor
Enclosures:
1) My Reply to your comments
2) Vinayak's Lectures on my proof.

Page 144
 Reply to Prof German Sierra Comments on RH

Dear Professor Sierra,

I thank you for comments on my paper that you have submitted to the Panel
of Scientists.
I wish to reply as follows:

1 Regarding your Comments

I believe you have a fundamental misunderstanding regarding the role of ran-
domness in my paper. You, rightly, point to several examples of previous work
where randomness was assumed and, again rightly, say that those were at best
conjectures which left their conclusions heuristic, which you say is the most
that can be said of my proposed proof. However, all those cases you cited were
mostly regarding the conjectured randomness of primes and its implications.
But all these matters play no role in my proof.
In fact I make no attempt to prove the randomness of primes (a conjecture
which is very dicult to prove and, luckily for me, is not required for the proof
of RH). What I use in my proof is, the randomness of the sequence of λ(n)
(which appears as the nth term in the Liouville series), in the sense that the
λ-sequence λ(n) = −1 or + 1 are
(series) resembles a random walk, in that (a)
equally probable and (b) λ(n) is not dependent on previous values of the λ's upto
any nite distance, as the length N of the sequence tends to ∞. This statistical
resemblence to a random walk for large N, persists even though some individual
members of the λ's are related by the deterministic relation λ(mn) = λ(m).λ(n).
These considerations, I then show is enough to prove RH, using the Khinchin
and Kolmogorov's Law of the Iterated Logarithm. Both (a) and (b) can be
proved in more than one way and I have done so in my paper. Actually, as
I have shown, the fact that λ's are deterministic and obey the deterministic
relation λ(m.n) = λ(m).λ(n), and therefore not random in the classical sense,
does not matter
1 at all for large N. If (a) and (b) can be proved then RH
is proved, as you yourself have agreed (see paragraph following Eq.(3) in your
review).
In your comments
2 you have mostly conned yourself to only a part of the
It was NOT
actual arguments that lead to the result I called Equal Probability.
my intention to say that Equal Probabilities is enough to prove RH. But this is
precisely what you seem to have thought because there is no reference by you to
1 For example, given λ(n) for some n, the formula λ(m.n) = λ(m).λ(n),can determine the
next predictable value λ(2.n) = λ(2).λ(n) = −λ(n), but since we consider very long sequences
(length N → ∞) for a large n, say n = 10100 , the integer 2n will be at a distance of 10100
from n making such a prediction statistically insignicant!
2I am well aware of the works you cited in your references, namely Borwein etal. your Ref
[3], Good and Churchhouse, your Ref [4] and Peter Humpries, your Ref [5] . While the rst
two deal mostly with the Mobius function and the third more with Polya's conjecture. (In
footnote 4, I briey state the diculties in using the Mobius function). None of them actually
PN
make a detailed study of of the growth of L(N) = n=1 λ(n) as N →∞ nor do they make a
rigorous comparision with a random walk (or coin tosses).

Page 145
the rest of the paper. I feel that this is an unfortunate misunderstanding, which I
wish to resolve, but I do, very much, need your kind patience and forebearance.
This is what was done in the rest of the paper. The purpose of this note is to
write out the rest of the Arguments.
Going back to Equal Probability. In the main paper I used a mapping and
pairing technique to obtain the result. I acknowledge that there are various
other prescriptions of mapping to prove the result. I also gave an Alternative
proof which does not use mapping but just pure induction (see my paper, Ref
[2], in Research gate). My intention was to tackle RH from rst principles,
this is the reason for my giving the two proofs.
In these proofs, while I focused on mapping, an alternate and equivalent
interpretation could be that I partition the number system into innite subsets
such that each natural number (except 1) can be uniquely assigned to only
one such subset. Each of these subsets have members whose λ values alternate
between +1 and −1 and therefore it is evident that its members have an equal
probability of having their λ values either +1 or−1. As any randomly chosen
natural number n will belong to one such subset, it will have an equal probability
of having λ(n) equal to +1 or −1,which is the Equal Probability result.
Yes, I too was very well aware that if one uses the Prime Number Theo-
rem and the fact that ζ(s) has no zero on the vertical line Re(s) = 1 and also
Littlewoods theorem (stated in my paper) it is possible to prove Equal Proba-
3
bilities see slide 238...in Ref [6] (below)Vinayak's lecture . I did not include this
because as the title of my paper has the phrase from rst principles.
However, as you have rightly pointed out that the Equal Probabil-
ity result is obtainable from the Prime Number Theorem. If so, you
have accepted that (a) is proved. What remains to prove is (b) that, the
λ(n) is independent of previous λ0 s (within a nite distance) in a long sequence
of length N → ∞. Therefore they would closely resemble the statistical behavior
of a ramdom walk (coin toss).
The independence of λ(n), for large arguments n,i.e. condition (b), is shown
in two dierent ways. In the rst (see Sec 11.3 Appendix IV of Ref.[1]) while it
is acknowledged that the λ0 s are linked through the multiplicative relationship
λ(mn) = λ(m).λ(n) etc., it is shown that the numbers linked in this manner
move increasingly further from each other so that this distance tends to innity
as n → ∞. Thus in the sequence {λ(n)} each λ(n) is independent of preceding
λ0 s which lie within a nite distance from it as n → ∞.
In the second proof of Independence (see para 2(a), p 21, in Ref[1]), it is
shown that any functional relationship which binds λ(n) to its previous λ0 s
(which lie within a distance L, arbitrary but xed) and which is valid for all
n,would make the sequence λ(n) periodic after some large n > n0 . It is then
proved that the sequence λ(n) cannot have such a cyclic behavior, because this
would imply, from Littlewood's theorem, that there are no zeros of the zeta
function within the critical strip - which is not true.

3 Vinayak Eswaran (my brother), had taken the trouble to write out 7 lectures on my proof
of RH, see Ref [6]. The lectures are well worth reading because of their clarity, lack of jargon
and assumes only the pre-reqisite of an under-graduate level of mathematical foundation.

Page 146
 With the properties (a) and (b) proved, a direct application of the Khinchine-
Kolmogorov Law of the Iterated Logarithm shows that the RH is TRUE and
that the width of the critical line is zero!

In the next section onwards I outline the main steps of the proof, which will
be a useful to read the paper. (I also strongly recommend Vinayak's Lectures
Ref [6] in the Reference Section (below), the lectures provide a detailed, and
sometimes an alternative, version of my proof ).
In the last I have added an experimental Verication which is not a part of
the proof. But the extensive calculations verify the properties of the lambda's
over very large sequences and hence lend support and credence to our methods.

2 Gist of The Proof (See Ref. 3 for Details)

The proof proceeds in essentially 4 basic steps.
STEP 1: We choose
4 an analytic function, F (s) = ζ(2s)/ζ(2s), whose poles
exactly correspond to the non-trivial zeros of the zeta function ζ(s). F (s) is
analytic in the region Re(s) > 1and is given by:

∞
X λ(n)
F (s) = (1)
n=1
ns

STEP 2: An analytical continuation of F (s) to the left of the vertical Re(s) = 1,

using Littlewood's Theorem determines that the asymptotic behavior of the
summatory function L(N ) :
N
X
L(N ) = λ(n) (2)
n=1

as N → ∞ plays a crucial role in determining the analyticity of F (s) and the

position of the poles of F(s) and thereby the zeros of ζ(s) in the critical region
0< Re(s) <1. Littlewood's theorem states that the asymptotic behaviour of
L(N ) for large N, determines the analyticity of F(s), and if the behaviour is
such that

N
X
| L(N ) | ≡ | λ(n) |< C N a+ (f or large N ) (3)
n=1

(where (1/2 ≤ a < 1 ), and  is a small positive number), F(s) will be analytic
in the region (a < Re(s). This is a very crucial result as far as RH is concerned
4 It may be mentioned here that for his study Littlewood had chosen the function 1/ζ(s)
which had lead to the µ-sequence. A diculty with using this Mobius function is that µ(n)
can take values−1, 0, or + 1, whereas λ(n) takes values of −1 or + 1, like a coin toss. Because
of this it is easier to compare the λ-sequence with coin tosses rather than the µ-sequence. And
because of this very reason I chose to study the function F (s) = ζ(2s)/ζ(s) , which leads to
the λ-sequence.

Page 147
because if one can determine that actually a = 1/2 in (3) then the Riemann
Hypothesis is proved.
STEP 3: In this step we logically put forth the argument: that the very
necessity that (3) must be satised for the Riemann Hypothesis to be true, im-
poses very severe resrtictions on the behaviour of the sequence of the Liouville
functions: {λ(1), λ(2), λ(3), .....}. These restrictions (conditions) are given later
in this section.
The λ(n) λ(1) = 1and for n > 1 : λ(n) = (−1)Ω(n) and
is dened as:
is determined by factorizing n and nding Ω(n), the number of prime factors
of n (multiplicities included). We already know λ(n) is fully determined by
factorizing n and is an arithmetic function namely: λ(m.n) = λ(m).λ(n), for all
integers m, n.
Now for RH to be true one must have a = 1/2 in Eq.(3), the rst N terms
(N large) of the λ sequence must therefore sum up as:

| λ(1) + λ(2) + λ(3) + ...... + λ(N ) | ' C . N 1/2 (4)

The above equation brings to mind a similar relationship satised by another

sequence c(n) = ±1, which corresponds to the nth step of a One-dimensional
random walk! (This c(n) can be simulated by coin tosses, if we replace Heads
by +1 and Tails by -1; so a N-step random walk can be thought as a coin toss
experiment where a coin is tossed N times.) It is well known that for such
a random-walker's sequence the sum indicates the distance travelled from the
starting position in N steps and satises the relationship:

| c(1) + c(2) + c(3) + ...... + c(N ) | ' C . N 1/2 (5)

The well known result of Equation (5), (see S.Chandrasekar), is derived by

using the assumption that the random walker behaves in such a manner that:
(i) Each step is of the same size but can be either in the positive direction or
negative direction i.e the nth step c(n) can be +1 or -1, with Equal Probability.
(ii) The sequence of steps cannot be periodic, that is the pattern of steps
cannot form a repetitive pattern (there is no cycle)
5.
th th
(iii) Knowing the n step the (n+1) cannot be predicted. That is, knowing
c(n), c(n+1) cannot be determined (they are independent).

The above assumptions are enough to derive Eq(6). This has been shown
by many researchers (e.g. See S. Chandrasekar, refered in Ref[1])

2.1 The Argument:

Comparing (4) and (5) leads us to deduce some inevitable conclusions:

Eq.(4) must be satised by the λ(n) sequence if the Riemann Hypothesis
is TRUE, this is the conclusion that we deduce from Littlewoods Theorem,
with the proviso that Eq. (4) needs be satised only for large N (this being

5 The requirement that there are no cycles was not necessary for Chandrasekar, but it is
necessary for the proof of (iii) i.e. independence.

Page 148
the condition of Littlewood's theorem). Now, there are many Random walks
possible, for instance: 100 random walkers can each of them, take N steps and
each of these random walkers will be at anapproximate distance of distance
C . N 1/2 from the starting point. Each of these 100 sequences can be thought of
as 100 dierent instances of a random walk of N steps each.
If we wish to compare (5) with (4) there are several conceptual issues: (α)
The sequence in (4) is a deterministic sequence and (β) we have only one se-
quence. We get over this latter issue by considering the single sequence as one
instance of a hypothetical random walk of N steps. And even though the λ(n)'s
are deterministic (an aspect we temporarily ignore) we could investigate this
one instance and argue (or hypothesize) that when N is large, the following
rules could be obeyed:
Properties of the λ−sequence
(a) Given an arbitrary large n chosen at random, there is Equal probability
of λ(n)being either +1 or-1.
(b) The λ−sequence cannot be periodic, that is the λ(n) cannot form a
repetitive pattern (no cycle)
(c) Knowing the value of λ(n) it is not possible to predict λ(n + 1) . Unpre-
dictability (independence).

Note the rules (a),(b)and (c) are similar to (i),(ii) and (iii) and therefore:
If by using the number theoretical (arithmetical) properties of the integers, the
primes and the factorization process, it is somehow possible to prove that the
λ(n) satisy the rules (a), (b) and (c) then just as (5) is satised by every instance
of a random walk, Eq(4) will be satised for our one particular instance of our
λ−sequence and thus RH will be proved! Taking this as a cue we proceed. NOTE:
According to Littlewood's Theorem: It is only necessary that the λ − sequence
satisy the rules for very large lengths of the sequence and large arguments of λ.
It is because of this relaxation provided by Littlewoods theorem that even though
the λ−sequence is deterministic, but its behaviour still very closely approximates
to the statistical behaviour of a sequence of random walks (or coin tosses) over
large N.
Hence the next step is to prove the properties for large values of N i.e. when
N tends to innity. (It will also become clear later that the deterministic nature
of the λ(n)0 s, does not signicantly disturb the above statistical properties. )
6
However, we are actually now at the crossroads: we have to prove that
the λ−sequence possesses the above properties (a),(b) and (c) for large sequence
lengths and large arguments N . Property (a) has already been proven, it is quite
possible that by using the artithmetic properties of the λ(n) that (b) can be
proved (as has been done in the Appendix III of the Main Paper) but it is in
the proving of (c) that the real diculty lies. This is because all mathematical
proofs in Arithmetic relies heavily on the Axioms of Peano (P.A), but P.A.
does not come to our aid for certain hard problems e.g to prove (or disprove)
that the advent of primes are random. Luckily we don't have to decide upon
this last surmise! But, we do have to decide upon the problem of resolving

6 See Foot note 1, on page 1.

Page 149
what is independence or unpredictability of a sequence. So we dene that
a sequence {a} ≡ {a(1), a(2), .....a(n), ....} as unpredictable, for large values of
its arguments, if when given the value of a(N ) where N is large and the M
previous values where M << N (and M nite), viz.{a(N − M ), a(N − M +
1), ....., a(N − 1), a(N )} then it is not possible to predict a(N + 1). It can be
easily seen that if a sequence {a} has this property then since, it is not possible
to predict the value of a(N + 1) knowing a(N ) and its M previous values, we
can assert that a(N ) and a(N + 1) are independent. It will be proved that by
this denition the λ−sequence ≡ {λ(1), λ(2), ....., λ(N ), ....}the components of
λ(N ) and λ(N + 1), are independent for large values of N. Using this knowledge
we can proceed.

STEP 4: Proof of the Properties of the λ−sequence. In this step several

theorems are proved using the number theoretical (arithmetical) properties of
integers, primes and the unique factorization of integers to establish the prop-
erties (a), (b) and (c) of the λ−sequence as listed in the previous paragraphs.
These proofs are fairly straight forward and are from rst principles:
We have already seen that Property (a) On Equal Probabilities, is proved
in Theorems 2 and 3 in Section 5.2, in the Main Paper Ref [1] and that the
concept of towers is used in the proofs. An alternative proof
7 by constuction
of all prime products and induction is also given in Ref [2]. A third proof, which
follows from Littlewood's theorem but assumes the fact that there is no zero
with Re(s)=1, (proved in the Prime Number Theorem) can also be derived as
has been discussed in the foregoing and also by you (but is not given in the
paper).
Property (b) On no cycles, is proved in Appendix III, Ref [1]
Property(c) On unpredictability (independence) is proved in Appendix IV,
Ref[1]. An alternative proof of this also given: See para 5(a), in page 2, of Ref
[3].
An alternative arithmetical proof of the asymptotic relation | L(N ) |' c =
C.N 1/2 . is given in Appendix V, Ref [1].
8

7 This alternative proof of equal probabilities is given in Ref [2] and is done by explict
construction of integers by products of sets which are powers of a given prime. Each of these
sets are seen to be comprised of members of ascending magnitudes and alternating λ−values,
in perfect analogy with the ascending and alternating odd/even sequences of the natural
numbers. Thus the member n of each set has the exact probability of 1/2 (as n → ∞) of its
λ−value being +1 or −1 as the natural numbers have of being odd or even. As every natural
number can be placed uniquely in one such subset, it is seen that a randomly chosen natural
number will have a probabilty of 1/2 of having its λ−value equal to +1 or −1(as n → ∞).
8 In a separate arithmetical study Ref [4], it was discovered that for very large N, smaller
primes contribute more (than the larger primes) to the calculation of λ(n)'s which occur in the
summatory expression for L(N ). Specically, if one chooses an integer K such that K << N
then the primes p which are s.t p < N/K , occur much more often in the calculation of each
term in L(N ) than the large primes q which are s.t. N/K < q < N . This situation permits
us to deduce, interestingly, that if we allow both K and N → ∞ in such a manner that the
CK
ratio N/K is a xed number, then we must have: P r(λ(n) = 1 | n < N ) = 1/2 − and
logN
CK
P r(λ(n) = −1 | n < N ) = 1/2 + logN where CK is a small uctuating number which tends
to zero as K → ∞; thus once again conrming that the L(N ) behaves like a random walk for
very large N.

Page 150
 We have therefore showed that Eq. (3) is satised by the λ−sequence. We
will now get an expression for the width of the critical line and show that this
width vanishes in the limit of large N.
We have established the fact that the λ−sequence behaves like a coin tosses
(or a random walk) and this entitles us to use Khinchine -Kolmogorov's law
of the Iterated Logarithm, adapted to the present context, is: Let {λn } be
independent, identically distributed random variables with means zero and unit
variances. Let SN = λ1 + λ2 + ... + λN . The limit superior (upper bound) of
SN almost surely (a.s.) satises
SN
Lim Sup √
2N loglogN
= 1 as N → ∞
Now, from Theorem 4 we have written that if we consider the λ0 s as coin
1
+dN
tosses one can write | L(N ) |=| λ(1) + λ(2) + ... + λ(N ) |≤ C0 N 2 (as
N → ∞) (since we are interested in only the behaviour for large N we henceforth
ignore the constants). Comparing this expression with the one above we see
1 √
that one can write N 2 +dN ∼ N loglogN thus yielding an expression for dN =
logloglogN
. We see that dN → 0 as N → ∞ , this satises the equivalent
2logN
L(N )
statement of (3) viz. 1 = 0, for any chosen >0 (Ref. Eq.(20) in Ref.[1])
N 2 +
. Further d∞ ≡ (LimN →∞ dN ) is the half-width of the critical line. Since this
is zero, we conclude that all the non- trivial zeros of the zeta function must lie
strictly on the critical line.
Thereby, the Riemann Hypothesis is proved.

3 Experimental verication
In the last Appendix VI, of the Main Paper, Ref[1], numerical experiments
(using Mathematica) are described and there it is shown that large sequence of
lambdas behave like a random walk (or equivalently like coin tosses).
I have calculated consecutive λ(n)0 s forming a large sequence, denoted by
Λ[N0 , M ], of length M of the form Λ[N0 , M ] ≡ {λ(N0 ), λ(N0 + 1), λ(N0 +
2), ..., λ(N0 + M − 1)}.where N0 is some large integer.
It has been shown by actually computation and performing a χ2 t using the
9
methods suggested by Donald Knuth, that these very large sequences contain-
ing consecutive values of λ(n)0 s very closely resemble and are in fact statistically
indistinguishable from a Binomial distribution (coin tosses) of equal length.
I have done very many computations (using Mathematica) and some of them
have been presented in the Tables in Appendix VI,sec.3, e.g. Tables 1.3 and
1.4 page 27; also see End Note on the last page of this document. These are
accurate and actual computations and the numerical results are indisputable
10 .
By very many numerical computations I have shown that the sets of consec-
p √ p
utive λ0 s denoted as S+ (N ) = Λ(N +1, N ) and S− (N ) = Λ(N − N +1, N )
, (N a square integer) have the property of being statistically indistinguishable
9 Knuth D.,(1968) `Art of Computer Programming', vol 2, Chap 3. Addison Wesley
10 I believe my papers provides the raison d'etre for the existence of this phenomena.

Page 151
from coin tosses.
11
These sequences called S− (N ) and S+ (N ) exist (N being a perfect square)
and behave like random sequences (coin tosses) and the concatination of such
sets of S− (N ) and S+ (N ) cover all of λ(n) for all integers n upto innity. This
shows that the entire {λ(n)} sequence is made up of an innite series of subse-
quences of type S− (N ) and S+ (N ) each of which statistically behave like coin
tosses! The Tables 1.4 cited above, provide ample proof of this. The purpose of
this section is just to demonstrate that the, predictions of the theorems proved in
the Main Paper, have been numerically veried extensively. The verications
12
2
have been done by doing a χ t of a λ−sequence with a Binomial distribution
(coin tosses). In every case it has been shown that for large N the λ−sequence
is indistinguishable from a random walk (sequence of coin tosses).
These numerical computations and χ2 correlations are very real and are ac-
tually present and give very strong indications of randomness present in the
λ(n)0 s which were actually computed by the factorization of integers n. I
strongly believe this phenomena has to be explained by the Pure Mathemati-
cians and not brushed aside or put under the carpet or carelessly labeled as mere
coincidence!
I wish to emphasize, that I have not only explained this phenomena but also
showed how it connects with the proof of the Riemann Hypothesis.

4 Conclusion
In this write up I have shown that the reason for the RH to be true lies with
the fact that the λ- sequence behaves statistically like coin tosses. It was shown
that a sequence c(n) of coin tosses or a sequence of a random walk, exhibits the
square root behavoir of Eq(4), was deduced by Khinchine-Kolmogorov, Chan-
drasekar and others, from the assumption of two criteria (i) Equal Probabilities
(ii) Independence. By using the properties of arithmetic and from the use of
mathematical deductions we could prove many theorems (and many have alter-
native proofs) to show that the λ- sequence for large values of its arguments
also satiisfy (i) and (ii). And therfore they satisfy Littlewood 's condition for

11 The reason for this was demonstrably argued because each integer n occurring in the
argument of λ(n) in one of the sets say S+ (N ) belongs to a dierent Tower.
Notice that if you choose N = j 2 then the union of the two sets :
S+ (j 2 + 1, j) ∪ S− ((j + 1)2 − j, j + 1) is nothing but the sequence
{λ(j 2 +1), λ(j 2 +2), λ(j 2 +3), ..., λ((j +1)2 )}. That is they cover all the λ0 s with arguments
between two consecutive perfect squares, j 2 to (j + 1)2 . Now if you choose N = (j + 1)2 you
can cover the next region between the perfect squares ((j + 1)2 + 1) to (j + 2)2 and therefore
you can capture all the regions between two consecutive perfect squares by concatinating such
sets all the way up to innity - basically covering all integers by the union of sets S− and
S+ right up to innity.
12 In fact I have proved (in Appendix VI of the Main Paper) that if you do a χ2 t of
a sequence of Pλ0 s of length N with
√ a sequence of coin tosses of equal length (using Knuth's
method) then N n=1 λ(m+n) = χ N for N large. If necessary, one may consult the extensive
numerical calculations done at the end of the last Appendix involving λ−sequences as long as
100,000 and argumets of λ as large as 10, 000,000,000!

Page 152
RH to be true. We also made extensive numerical computations involving the
λ−sequence; all these support the various theorems we have proved. Since,
we have started from rst principles and used only the arithmetic properties
of numbers and mathematical logic to prove the theorems, in my opinion, this
leaves hardly any doubt as to the truth of RH.
My appeal to you: The RH is the greatest open problem in Mathematics.
I believe that I have solved it comprehensively. I request you to look at the
proof again (starting with the reading of Vinayak's lectures which can be done
quickly), as I found your review had missed many aspects which I believe, are
very crucial to the proof, (I beg your pardon for saying so!). This onerous task,
which I am requesting you to kindly take up, will not only do justice to my work
but also to the great problem it addresses. If you have any further queries or
doubts, it will be my pleasure to answer them.
In the End, as in all matters of consequence, Truth will prevail.
Thank You. And kind Regards
K. Eswaran

5 References
[1] The nal and Exhaustive proof of the Remann Hypothesis...

[2]A Simple Proof That Even and Odd Numbers of Prime Factors Occur
with Equal Probabilities in the Factor-ization of Integers

[3]A Quick Reading Guide to the Proof of the Riemann Hypothesis

[4] The eect of of the non-random-walk behavior of the Liouville Series

L(N) by the rst nite number of terms.

I enclose below the slides of the Invited Lecture that I delivered at the
Government Arts & Science College Kumbakonam on March 1st 2019. (This
was followed by another (slightly shorter) Lecture delivered in the Ramnujan
Centre of Sastra University on the evening of the same day).
[5] Invited Lecture On the Riemann Hypothesis by K.Eswaran

[6] Vinayak Eswaran: Seven Lectures of Kumar Eswaran's Proof on RH

K. Eswaran/Professor
Sree Nidhi Institute of Science and Technology,
Yamnampet, Ghatkesar, Hyderabad 501301
5th October 2020

6 END NOTES -These are just end notes which

-provide some more information.
Just for curiosity I tested the behaviour of very long sequences of lambda and
compared them with coin tosses by χ2 tting.See the Appendix VI of my Main
Paper Ref[1], there are many more examples in the form of Tables.

Page 153
 PK+L √
Here we dene the summation: SUM ≡ n=K λ(n) = χ L (see Eq(9), in
page 25 of Ref[1]).
Everywhere the χ2 ts get better and better as the the length of the sequence
and the size of the integer increases. Knuth speculated that a value of around
4 or 5 to make the sequence indistinguishable from a sequence of coin tosses or
a random-walk. However, Littlewood's criterion is far less strict, it is sucient
PK+N √
that as N tends to innity: n=K λ(n) = C N , where C can be any nite
value.
In the tests below the lambda sequence passes the test in every case. I
have taken very Long sequences, for Example III, I have considered 100,000
consecutive integers starting from K = 25 × 1024 + 1
EXAMPLE I: A sequence of Length L of consecutive lambdas starting from
λ(25000001) to λ(25005000) of 5000 λ values for 5000 consecutive integers,starting
from 25, 000, 001 ie L = 5000. We use Mathematica commands in our compu-
tations as shown below.
Plus[LiouvilleLambda[Range[25000001, 25005000]]]
= {1, 1, 1, 1, 1, 1, 1, -1, -1, -1, 1, -1, -1, 1, 1, 1, -1, -1, 1, 1, \ 1, -1, -1, 1, 1,
-1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, 1, -1, 1, -1, \ -1, -1, 1, -1, -1, -1, -1, 1, -1, -1, 1,
-1, 1, -1, 1, -1, 1, 1, 1, \TRUNCATED-- \ 1, -1, 1, 1, 1, 1, 1, 1, -1}
SUM= -42, χ2 = [SUM * SUM/ L] = [42*42/5000}= 0.3528

EXAMPLE II
Starting Integer K = 25 × 1024 + 1; L = 10, 000
Plus[LiouvilleLambda[ Range[25000000000000000000000001, 25000000000000000000010000]]]
SUM= 88, χ2 = 88*88/10000 = 0.7744

EXAMPLE III:
Starting Integer K = 25 × 1024 + 1; L = 100, 000
Plus[LiouvilleLambda[ Range[25000000000000000000000001, 25000000000000000000100000]]]
SUM= 238, χ2 = 238*238/ 100000 = 0.56644

EXAMPLE IV:
Starting Integer Z = 1030 + 1; L=1000;
Plus[LiouvilleLambda[Range[Z, Z + 999]]]
SUM= -20, χ2 = 20*20/1000 =0.4

EXAMPLE VI:
Starting Integer Z = 1030 + 1; L=10,000;
Plus[LiouvilleLambda[Range[Z, Z + 9999]]]
SUM= 54 χ2 = 54*54/10000 = 0.2916

10

Page 154
Part B: K Eswaran All Main Papers on RH
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/325035649

The Final and Exhaustive Proof of the Riemann Hypothesis from First
Principles

Preprint · May 2018

DOI: 10.13140/RG.2.2.35243.95528

CITATIONS READS
0 6,313

1 author:

K. Eswaran
Sreenidhi Institute of Science & Technology
87 PUBLICATIONS   325 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The Dirichlet Series for the Liouville Function and the Riemann Hypothesis View project

Development of New Pattern Recognition Algorithms for use in classification and Neural network Research View project

All content following this page was uploaded by K. Eswaran on 09 May 2018.

The user has requested enhancement of the downloaded file. Page 155
Eswaran, K.. (MAY 2018) “The Final and Exhaustive Proof of the Riemann Hypothesis from First Principles,”
Research Gate, Vol. MAY 2018, Article ID Version - FINAL
doi of Reseachgate/PREPRINT/Version - FINAL

The Final and Exhaustive Proof of the Riemann Hypothesis from First Principles

K.Eswaran1
1 Sreenidhi Inst. of Science and Technology, Yamnampet, Ghatkesar,Hyderabad 501301, INDIA

Correspondence to be sent to: keswaran@sreenidhi.edu.in

ABSTRACT: As is well-known, the celebrated Riemann Hypothesis (RH) is the prediction that all the non-trivial
zeros of the zeta function ζ(s) lie on a vertical line in the complex s-plane at Re(s) = 1/2. Very many efforts to prove
this statement have been directed to investigating the analytic properties of the zeta function, however all these efforts
have not been able to substantially improve on Riemann’s initial discovery: that all the non trivial zeros lie in verical
strip of unit width whose centre is the critical line. The efforts have been rendered difficult because of a lack of a
suitable functional representation (formula) for ζ(s) (or 1/ζ(s)) , which is valid and analytic over all regions of the
Argand plane; these difficulties are further complicated by the presence of prime numbers in the very definition of the
zeta function and the lack of predictability in the behaviour of prime numbers which makes the analysis intractable. In
this paper we make our first headway by looking at the analyticity of the function F (s) = ζ(2s)/ζ(s) that has poles in
exactly those positions where ζ(s) has a non trivial zero. Further, the trivial zeros of the zeta function, which occur at
the negative even integers, conveniently cancel out in F (s) and do not appear as poles of the latter (however there is an
isolated pole of F (s), viz. s = 1/2, which is actually a pole of ζ(2s) but this will not worry us because it is on the
critical line). So the task of proving the RH is some what ‘simplified’ because all we have to show is: All the poles of
F (s) occur on the critical line, which then is the main aim of this paper. We then investigate the Dirichlet series that
obtains from the function F(s) and employ novel methods of summing the series by casting it as an infinite number of
sums over sub-series. In this procedure, which heavily invokes the prime factorization theorem, each sub-series has the
property that it oscillates in a predictable fashion, rendering the analytic properties of the function F (s) determinable.
With the methods developed in the paper many theorems are proved, for example we prove: that for every integer with
an even number of primes in its factorization, there is another integer that has an odd number of primes (multiplicity
counted) in its factorization; by this demonstration, and by the proof of several other theorems, a similarity between
the factorization sequence involving (Liouville’s multiplicative functions) and a sequence of coin tosses is
mathematically established. Consequently, by placing this similarity on a firm foundation, one is then empowered to
demonstrate, that Littlewood’s(1912) sufficiency condition involving Liouville’s summatory function, L(N ), is satisfied.
It is thus proved that the function F (s) is analytic over the two half-planes Re(s) > 1/2 and Re(s) < 1/2, clearly
revealing that all the nontrivial zeros of the Riemann zeta function are placed on the critical line Re(s) = 1/2.

Extended Abstract
The paper approaches∗ the RH in the following way:

(1) This proof of the Riemann Hypothesis (Riemann 1859) crucially depends on showing that the function
F (s) ≡ ζ(2s)/ζ(s), has poles only on the critical line s = 1/2 + iy, which translates to having the non-trivial
zeros of the ζ(s) function on the self-same critical line. It can be easily verified that all the non-trivial zeros of
ζ(s) appear as poles in F (s), and all the trivial zeros cancel and so do not appear as poles in F (s)† . It can also
be proved, from symmetry considerations, that both the numerator and denominator of F (s) cannot vanish at
the same point. Hence, to prove the RH, all we need to show is that all the poles of F (s) occur on the critical line.

(2) A method applied by Littlewood (1912, see Edwards (1974) pp 260) to obtain an equivalent statement
of RH involving the 1/ζ(s) function is applied here to F (s) to obtain a previously-known equivalent statement of

Received 9 MAY 2018

Communicated by XXX
∗ This paper is the Final Version of the proof on RH, the earlier versions bear the title ‘The Dirichlet Series for
the Liouville Function and the Riemann Hypothesis’ Sept 2016 to Oct 2017) are listed in Ref. [8] K. Eswaran
† except that the latter has an extra pole on the critical line

c The Author MAY 2018. ResearchGate. All rights reserved contact Author for permissions,
please e-mail: journals.permissions@oxfordjournals.org.

Page 156
2 K. Eswaran

uncertain provenance. Littlewood’s method lies in analytically continuing the Dirichlet series for F (s), strictly
valid for Re(s) > 1, to the region Re(s) ≤ 1. The analyticity of F (s) turns out to be crucially dependent on the
boundedness of L(N )N −s as N → ∞, where L(N ) is the summatory Liouville function. That is, if L(N ) ∼ N a
(a > 0) asymptotically as N → ∞, then the F (s) can be analytically continued only on the right of the vertical
line drawn at Re(s) = a. In other words, the singularities of F(s) will lie on or to the left of Re(s) = a. Further,
since the non-trivial zeros of the Riemann zeta function ζ(s) exactly correspond to the poles of F(s), and are
known to be symmetrically placed about the Re(s) = 1/2 line, this automatically implies that the non-trivial
zeros of the zeta function will all be within the region 1 − a ≤ Re(s) ≤ a. If a → 1/2 from the right, the zeros
1
will lie on the critical line and RH will be true. This also means that L(N ) ∼ N 2 as N → ∞, thereby yielding
the RH-equivalent statement (see Borwein et al 2006, p. 48):

L(N )
limN →∞ 1 =0 for  > 0
N 2 +
PN
Here L(N ) = 1 λ(n), and λ(n) is the Liouville function that is +1 or -1 depending on whether n has an even
or odd number of prime factors (multiplicity included).
1
(3) The equivalent statement above requires, if the RH is true, that |L(N )| ∼ N 2 as N → ∞ (a < 1/2 is
impossible as the Re(s) = 1/2 line is known to have numerous zeros of the the zeta function). This expression
is strongly suggestive of X(N ), the distance travelled in N unit steps in a standard random walk, which can be
represented as:
XN
X(N ) = c(n)
n=1

where the c(n)’s are “coin-tosses”, i.e., independent random numbers with an equal probability of being either
+1 or -1. It is well-known that the expected value of |X(N )|, for large N , is

limN →∞ E(|X(N )|) = C0 N 1/2

1
Therefore, if it could be shown that the L(N ) series is a random-walk, and that |L(N )| ∼ N 2 as N → ∞, the
RH would be proved. This is the approach taken here. So we have to prove that the λ(n)’s in the L(N ) series
are essentially “coin-tosses”, for large n.

To show that the λ’s behave as coin tosses, we have to show that (i) their probabilities of being either +1
of -1 are equal, and further (ii) that the λ’s appearing in the natural sequence, n = 1, 2, 3, ..., are independent
of each other.

(4) Equal Probabilities: A crucial advance in this line of attack is the discovery of a method of factorizing
every integer and placing it in an exclusive subset, where it and its other members in the same subset form
an increasing sequence of natural numbers that alternately have odd and even numbers of prime factors. Such
subsets are called ‘Towers’ in the Paper. It is shown that every natural number, other than 1, has a unique place
in a unique Tower of ordered countably infinite members, and each such member represents a unique natural
number. The alternating odd and even factorization of the members of the Towers ensures that each Tower is
partitioned into equal proportions of members with λ’s of -1 and +1. As the entire natural number system is
incorporated within the Towers system, the natural numbers also have equal proportions with odd and even
numbers of prime factors, respectively with λ’s of -1 and +1. In other words, concerning the probability that
a random natural number n has a λ of either value, we conclude: P rob[λ(n) = +1] = P rob[λ(n) = −1] = 1/2
[Theorem 3B]‡ .

(5) Non-periodicity and Independence: We then show that the sequence of λ’s in L(N ) can never be cyclic,
just as a sequence of coin tosses can never be cyclic. This done in Appendix III and follows directly from
Littlewood’s method. Quasi-cyclicity or any other pattern of λ’s in the L(N ) series that would keep L(N )
bounded as N → ∞ are also excluded. Specifically, non-cyclicity would preclude any dependence of the type

λn = f (λn−1 , λn−2 , λn−3 , ..., λn−M )

‡ Theorems 3 and 3B, that argue for equal proportions and hence equal probabilities of the the λ’s over the natural number system,
were later given a new and different proof in K. Eswaran, (April 2018)

Page 157
 A Proof of RH 3

for finite M . So, essentially, the independence of λ’s in the natural sequence is proved [although within the
Towers, they have a perfectly predictable and deterministic relation]. Here we adopt the notation: λn for λ(n).

(6) Arithmetic Independence: The independence of the λ’s in the L(N ) series mentioned above followed
from Littlewood’s method, i.e., Analysis. In Appendix IV, a purely arithmetic approach is taken. It is shown
that from merely these two rules: λ(p) = −1 for a prime p, and λ(pq) = λ(p) × λ(q), where q is any integer, the
entire sequence of λ’s for n = 1, 2, 3, ..., can be obtained, in a manner reminiscent of the construction of the
natural number system by multiplication. It is then argued that for any two integers n > m, λ(n) is dependent
on λ(m), if the latter is required to find the former, and independent if not. It is then shown that, as n → ∞,
any finite sequential strip of λ’s will be independent of each other, thus essentially making them equivalent to
coin-tosses.

(7) With Theorem 3B and Appendices III and IV, we have proved that the L(N ) series is a random walk
of infinite length. We then invoke (towards the end of Section 5) Khinchin and Kolmogorov’s law of the iterated
logarithm to show that the maximum deviation, from the one-half power-law expectation, in the exponent of
1
|X(N )| for any individual random walk tends monotonically to zero as N → ∞. So, in fact, |L(N )| ∼ N 2 as
N → ∞. Therefore, for any chosen  > 0, the equivalent statement for the validity of RH will be satisfied and
the Riemann Hypothesis is proved.

Interestingly, the aforementioned maximum deviation from the one-half power-law expectation in the
random walk of L(N ) is exactly the half-width of the critical strip around Re(s) = 1/2 that contains all the
zeros of the of the zeta function. As that deviation approaches zero as N → ∞, that width is also zero, ensuring
that all the non-trivial zeros of the zeta function lie on Re(s) = 1/2.

(8) In Appendix V, starting from Littlewood’s ansatz, that |L(N )| ∼ N a , for N → ∞, we argue that
the statistics of the λ’s must become “self-similar” over large consecutive sequences of λ’s. It is shown that,
if we choose two sets S− (N ) and S+ (N ) of consecutive integers, each of them containing k integers,§ then
the λ’s defined over these sets S− (N ) and S+ (N ) are statistically similar to each other. This statistical
similarity is shown to hold for all large k i.e for all N = k 2 , which implies the statistical behavior of λ’s are
independent of the length k of S− (N ) and S+ (N ) for large but arbitrary N = k 2 . This principle yields us
the value of a = 21 , which again would satisfy the equivalent statement of the RH. This ‘physicist’s proof’
of the RH, is separate from the argument in the main paper, and may be treated as an interesting addendum to it.

(9) Finally, in Appendix VI, we show the sequence of λ’s is statistically indistinguishable from coin tosses
(using the χ2 statistical test) over many sets of consecutive integers (as was demonstrated in Appendix V).
Further it is also shown that the sequence λ’s is indistinguishable from coin tosses over the entire range of
numbers from n = 1 to 176 trillion. This verification has been done by actual numerical computation over large
sets of integers (which are below 176 trillion). While this is merely a ‘verification’, not a ‘proof’, this empirical
result follows directly from our proof of the Riemann Hypothesis, and affirms its sound basis.

Interestingly, it is also shown in Appendix VI that a connection exist between L(N ) and χ2 when compared
to a sequence of coin tosses. From the relation Eq.(9), p 24, one can conclude that to satisfy Littlewood’s
condition, only the first and second moments of the distributions of lambda and coin toss sequences need be
similar.
Because of the extensive computations and calculations made, which are backed by theory, Appendix VI
can be thought of as an experimental physicists’ verification of the ‘Law of Riemann’.

1 Introduction
This paper investigates the behaviour of the Liouville function, (ref. Apostol (1998), which is related to Riemann’s
zeta function, ζ(s), defined by
∞
X 1
ζ(s) = , (1.1)
ns
n=1

where n is a positive integer and s is a complex number, with the series being convergent for Re(s) > 1.
This function has zeros (referred to as the trivial zeros) at the negative even integers −2, −4, . . .. It has been

§ The
set S− (N ) contains k consecutive integers ending with the integer N = k2 , and S+ (N ) contains the next k consecutive
numbers, see Appendix V for details and definitions

Page 158
4 K. Eswaran

shown¶ that there are an infinite number of zeros on the line at Re(s) = 1/2. Riemann’s Hypothesis (R.H.)
claims that these are all the nontrivial zeros of the zeta function. The R.H. has eluded proof to date, and
this paper demonstrates that it is resolvable by tackling the Liouville function’s Dirichlet series generated by
F (s) ≡ ζ(2s)/ζ(s), which is readily rendered in the form
∞
X λ(n)
F (s) = , (1.2)
ns
n=1

where λ(n) is the Liouville function defined by λ(n) = (−1)Ω(n) , with Ω(n) being the total number of prime
numbers in the factorization of n, including the multiplicity of the primes. We would also need the summatory
function L(N ), which is defined as the partial sum up to N terms of the following series:

N
X
L(N ) = λ(n) (1.2b)
n=1

Since the function F (s) will exhibit poles at the zeros of ζ(s), we seek to identify where ζ(s) can have zeros
by examining the region over which F (s) is analytic. By demonstrating that a sufficient condition, derived by
Littlewood (1912),(in Edwards(1974)), for the R.H. to be true is indeed satisfied, we show that all the nontrivial
zeros of the zeta function occur on the ‘critical line’ Re(s) = 1/2.
Briefly, our method consists in judiciously partitioning the set of positive integers (except 1) into infinite
subsets and couching the infinite sum in (1.2) into sums over these subsets with each resulting sub-series being
uniformly convergent. This method of considering a slowly converging series as a sum of many sub-series was
previously used by the author in problems where Neumann series were involved Eswaran (1990)).
In this paper we break up the sum of the Liouville function into sums over many sub-series whose behaviour
is predictable. It so turns out that one prime number p (and its powers) which is associated with a particular
sub-series controls the behaviour of that sub-series.
Each sub-series is in the form of rectangular functions (waves) of unit amplitude but ever increasing
periodicity and widths - we call these ‘harmonics’ - so that every prime number is thus associated with such
harmonic rectangular functions which then play a role in contributing to the value of L(N ). It so turns out that
if N goes from N to N + 1, the new value of L(N + 1) depends solely on the factorization of N + 1, and the
particular harmonic that contributes to the change in L(N ) is completely determined by this factorization. Since
prime factorizations of numbers are uncorrelated, we deduce that the statistical distribution of L(N ) when N
is large is like that of the cumulative sum of N coin tosses, (a head contributing +1 and a tail contributing -1),
and thus logically lead to the final conclusion of this paper.
We found a new method of factoring every integer and placing it in an exclusive subset, where it and its
other members form an increasing sequence which in turn factorize alternately into odd and even factors; this
method exploited the inherent symmetries of the problem and was very useful in the present context. Once this
symmetry was recognized, we saw that it was natural to invoke it in the manner in which the sum in (1.2) was
performed. We may view the sum as one over subsets of series that exhibit convergence even outside the domain
of the half-plane Re(s) > 1. We were rewarded, for following the procedure pursued in this paper, with the
revelation that the Liouville function (and therefore the zeta function) is controlled by innumerable rectangular
harmonic functions whose form and content are now precisely known and each of which is associated with a
prime number and all prime numbers play their due role. And in fact all harmonic functions associated with
prime numbers below or equal to a particular value N determine L(N ). The underlying symmetry being alluded
to here, remained hidden because the summation in (1.2) is written in the usual manner, setting n = 1, 2, 3, ...
in sequence.
From the next section onwards the paper follows the plan enunciated in the Extended Abstract and indicated
by the the steps (1) to (9) detailed therein.

2 Partitioning the Positive Integers into Sets

The Liouville function λ(n) is defined over the set of positive integers n as λ(n) = (−1)Ω(n) , where Ω(n) is the
number of prime factors of n, multiplicities included. Thus λ(n) = 1 when n has an even number of prime factors
and λ(n) = −1 when it has an odd number of prime factors. We define λ(1) = 1. It is a completely arithmetical
function obeying λ(mn) = λ(m)λ(n) for any two positive integers m , n.

¶ This was first proved by Hardy (1914).

Page 159
 A Proof of RH 5

We shall consider subsets of positive integers such as {n1 , n2 , n3 , n4 , ...} arranged in increasing order and
are such that their values of λ alternate in sign:

λ(n1 ) = −λ(n2 ) = λ(n3 ) = −λ(n4 ) = ... (2.3)

It turns out that we can label such subsets with a triad of integers, which we now proceed to do. To construct
such a labeling scheme, consider an example of an integer n that can be uniquely factored into primes as follows:

n = pe11 pe22 pe33 ...peLL pi pj (2.4)

where p1 < p2 < p3 ... < pL < pi < pj are prime numbers and the ek , k ∈ {1, 2, 3, ..., L} are the integer exponents
of the respective primes, and pL is the largest prime with exponent exceeding 1, the primes appearing after pL
will have an exponent of only one and there may a finite number of them, though only two are shown above.
Integers of this sort, with at least one multiple prime factor are referred to here as Class I integers. In contrast,
we shall refer to integers with no multiple prime factors as Class II integers. A typical integer, q, of Class II may
be written
q = p1 p2 p3 ...pj pL , (2.5)
where, once again, the prime factors are written in increasing order.
We now show how we construct a labeling scheme for integer sets that exhibit the property in (2.3) of
alternating signs in their corresponding λ’s. First consider Class I integers. With reference to (2.4), we define
integers m, p, u as follows:
eL−1
m = pe11 pe22 pe33 ...pL−1 ; p = pL ; u = pi pj . (2.6)
In (2.6), m is the product of all primes less than pL ,the largest multiple prime in the factorization, and u is the
product of all prime numbers larger than pL in the factorization. Thus the Class I integer n can be written

n = mpeL u (2.7)

Hence we will label this integer n as (m, peL , u),using the triad of numbers(m, p, u) and the exponent eL . It is
to be noted that u will consist of prime factors all larger than p, and u cannot be divided by the square of a
prime number.
Consider the infinite set of integers, Pm;p;u , defined by

Pm;p;u = {mp2 u, mp3 u, mp4 u, ...} (2.8)

The Class I integer n necessarily belongs to the above set because eL ≥ 2. Since the consecutive integer members
of this set have been obtained by multiplying by p, thereby increasing the number of primes by one, this set
satisfies property (2.3) of alternating signs of the corresponding λ’s. Note that the lowest integer of this set
Pm;p;u of Class I integers is mp2 u.
We may similarly form a series for Class II integers. The integer q in (2.5) may be written q = mpu, with
m = p1 p2 p3 ...pj , p = pL , and u = 1. This Class II integer is put into the set Pm;p;u defined by

Pm;p;1 = {mp, mp2 , mp3 , mp4 , ...}. (2.9)

The set containing Class II integers is distinguished by the facts that u = 1 for all of them, their largest prime
factor is always p and none of them can be divided by the square π 2 of a prime number π such that π < p; in
other words the factor m cannot be divided by the square of a prime. In this set, too, the λ’s alternate in sign
as we move through it and so property (2.3) is satisfied. Again, note that the lowest integer of this set Pm;p;1 is
the Class II integer mp, all the others being Class I.
In what follows, we shall find it handy to refer to the set of ascending integers comprising Pm;p;u as a ‘tower’.
It is important to distinguish between a tower (or set) described by a triad like (m, p, u) and an integer belonging
to that set. It is worth repeating that the set or tower of Class I integers described by the label (m, p, u) is the
infinite sequence {mp2 u, mp3 u, mp4 u, ...}, the first element of which is mp2 u and all other members of which are
mpk u, where k > 2. A set or tower containing a Class II integer described by (m, p, u = 1) is the infinite sequence
{mp, mp2 , mp3 , ...}, Eq.(2.9), of which only the first element mp is a Class II integer and all other members,
mpk , where k ≥ 2, are Class I, because the latter have exponents greater than 1. For convenient reference, we
shall refer to the first member of a tower as the base integer or the base of the tower. It is also worth noting that
when we refer to a triad like (m, pk , u), where k > 1, we are invariably referring to the integer mpk u and not to
any set or tower. Labels for sets do not contain exponents; only those for integers do. Of course, the particular
integer (m, pk , u) belongs to the set or tower (m, p, u).

Page 160
6 K. Eswaran

Two simple examples illustrate the construction of the sets denoted by Pm;p;u :
Ex. 1: The integer 2160, which factorizes as 24 × 33 × 5, is clearly a Class I integer since it is divisible by the
square of a prime number—in fact there are two such numbers, 2 and 3—but we identify p with 3 as it is the
larger prime. It is a member of the set P16;3;5 = {16 × 32 × 5, 16 × 33 × 5, 16 × 34 × 5, 16 × 35 × 5, ...}.
Ex. 2: The integer 663, which factorizes as 3 × 13 × 17, is a Class II integer because it is not divisible by the
square of a prime number. It belongs to the set P39;17;1 = {39 × 17, 39 × 172 , 39 × 173 , ....}.
Note that two different integers cannot share the same triad.k And two different triads cannot represent
the same integer.∗∗ Thus the mapping from a triad to an integer is one-one and onto. A formal proof is in the
Appendix.
The following properties of the sets Pm;p;u may be noted:
(a) The factorization of an integer n immediately determines whether it is a Class I or a Class II type of
integer.
(b) The factorization of integer n also identifies the set Pm;p;u to which n is assigned.
(c) The procedure defines all the other integers that belong to the same set as a given integer.
(d) Every integer belongs to some set Pm;p;u (allowing for the possibility that u = 1) and only to one set.
This ensures that, collectively, the infinite number of sets of the form Pm;p;u exactly reproduce the set of positive
integers {1, 2, 3, 4, ....}, without omissions or duplications.
Our procedure, taking its cue from the deep connection between the zeta function and prime numbers, has
constructed a labeling scheme that relies on the unique factorisation of integers into primes. In what follows, we
shall recast the summation in (1.2) into one over the sets Pm;p;u .The advantage of breaking up the infinite sum
over all positive integers into sums over the Pm;p;u sets will soon become clear.

3 An alternative expression of the Liouville Function’s Dirichlet series

The usefulness of this Section and the next (i.e. Sections 3 and 4) is to show that the cumulative summatory
function L(N ) = ΣN n=1 λ(n), can be built up by ‘harmonic rectangular waves’, thus providing a pictorial
representation of the function L(N ). This pictorial view which helped us to understand the phenomena in
RH, actually followed the discovery of an alternative expression for Eq(1.2) namely the representation Eq(3.10).
This expression is written in terms of ‘towers’ which as we shall see help in our study of the properties of L(N ).
The kinks in the rectangular waves which occur at integer values k in the argument of λ(k) each contribute
either +1 or -1 to L(N ) and are distributed like coin tosses and their summation is akin to the cumulative sum
of N coin tosses. Eq. (3.10) has also helped in evolving the concepts of Towers described in the section preceding,
and apart from this Eq(3.10) plays no crucial role in the proof of RH.

THEREFORE ON A FIRST READING THE FOLLOWING CAN BE OMITTED: SEE FOOT NOTE.††

We shall now implement the above partitioning of the set of all positive integers to examine the analytic
properties of F (s) in (1.2). We shall rewrite the sum in (1.2) into an infinite number of sums of sub-series.
We begin, however, by assuming that Re(s) > 1, which makes the series in (1.2) absolutely convergent, in
fact it represents ζ(2s)/ζ(s) and is well defined for Re(s) > 1. We will not be needing the expression for regions
Re(s) < 1. ‡‡ We write the right hand side in sufficient detail so that the implementation of the partitioning

k The integer represented by the triad (m, pr , u), is the product mpr u, which obviously cannot take on two distinct values.
∗∗ Suppose two different triads (m, pr , u) and (µ, π ρ , ν) represent the same integer, say n. Then we must have mpr u = µπ ρ ν = n.It
follows that at least two numbers of the tetrad {m, p, r, u} must differ from their counterparts in the tetrad {µ, π, ρ, ν}. Since the
factorization of n is unique, this is impossible.
†† On a first reading Sections 3 and 4 can be omitted. And one can go directly to Section 5 and after reading the proof of Theorem
1, in Section 5.1, skip the rest of this subsection and go directly to subsection 5.2, in page 12 and read till the end of the paper.
Though, the last four paragraphs of Section 4 should be read to understand the Figure. Sections 3 and 4 have been included to
maintain mathematical rigor: To demonstrate that expressions (1.2) and (3.10) have the same analytical continuation to the left of
Re(s) = 1 by Littlewood’s theorem.
‡‡ Though each sub-series is convergent for Re(s) < 1- see Titchmarsh or Whittaker and Watson.

Page 161
 A Proof of RH 7

scheme becomes self-evident:

∞ ∞ ∞ ∞
X λ(2r ) X λ(3r ) X λ(5r ) X λ(2 × 3r )
F (s) = 1+ + + +
2rs 3rs 5rs 2s 3rs
r=1 r=1 r=1 r=1
∞ ∞ ∞ ∞
X λ(7r ) X λ(2 × 5r ) X λ(11r ) X λ(2k × 3)
+ + + +
7rs 2s 5rs 11rs 2ks 3s
r=1 r=1 r=1 k=2
∞ ∞ ∞ ∞
X λ(13r ) X λ(2 × 7 ) r X λ(3 × 5 ) X λ(17r )
r
+ + + +
13rs 2s 7rs 3s 5rs 17rs
r=1 r=1 r=1 r=1
∞ r ∞ ∞ ∞
X λ(19 ) X λ(2k × 5) X λ(3 × 7 ) r X λ(2 × 11r )
+ + + +
19rs 2ks 5s 3s 7rs 2s 11rs
r=1 k=2 r=1 r=1
∞ r ∞ r ∞ k ∞
X λ(23 ) X λ(2 × 13 ) X λ(2 × 7) X λ(29r )
+ + + +
23rs 2s 13rs 2ks 7s 29rs
r=1 r=1 k=2 r=1
∞
X λ(2 × 3 × 5r )
+ + ··· , (3.10)
2s 3s 5rs
r=1

We have explicitly written out a sufficient number of terms of the right hand side of (1.2) so that those
corresponding to each of the first 30 integers are clearly visible as a term is included in one (and only one) of
the sub-series sums in (3.10). On the right hand side, the second term contains the integers 2, 4, 8, 16....; the
third contains 3, 9, 27, ...; the fourth contains 5, 25, 125, ...; the fifth contains 6, 18, 54, ...; sixth contains 7, 49, ...;
the seventh contains 10, 50, ...; the eighth contains 11, 121, ...; the ninth contains 12, 24, 48, ...; and so on. Note
that in the ninth, fifteenth, and twentieth terms the running index is deliberately switched from r to k to alert
the reader to the fact that the summation starts from 2 and not from 1 as in all the other sums. (Note that, in
the ninth term, the Class I integer n = 12 = 22 × 3 is assigned to the set P1;2;3 = {22 × 3, 23 × 3, 24 × 3, ...} and
not to the set P4;3;1 = {22 × 3, 22 × 32 , 22 × 33 , .}, because the first term identifies p as 2 and u as 3 where as
the second term onwards 3 has exponents, which violates our rules of precedence and would be an illegitimate
assignment given our partitioning rules.)
The sub-series in (3.10) have one of two general forms:
∞
X λ(m.pr ) λ(m.p) 1 1 1 (−1)X
= [1 − + − + · · · + + ···]
ms .prs ms .ps ps p2s p3s pXs
r=1

or
∞
X λ(m.pk .u) λ(m.p2 .u) 1 1 1 (−1)X
= [1 − + − + · · · + + ···] (3.11)
ms .pk .us ms .p2s .us ps p2s p3s pXs
k=2

The above geometric series occurring within square brackets in the above two equations can actually be summed
(because they are convergent),(see Whittaker and Watson) but we will refrain from doing so, and (1.2) can be
rewritten as
XXX XX
T T
F (s) = Fm;p;u (s) + Fm;p;1 (s), (3.12)
m p u m p

where the first group of summations pertains to Class I integers n characterized by the triad (m, pk , u), (k ≥ 2)
and the second group pertain to those integers which are characterized by set (m, pk , 1), (k ≥ 1) the first member
in the set is a Class II integer and others Class I.
T
In the above we have defined the function Fm;p;u (s) of the complex variable s which is a sub-series involving
terms over only the tower (m, p, u) for a Class I integer as follows
∞
T
X λ(mpk u)
Fm;p;u (s) = , (3.13)
ms pks us
k=2

T
and the function Fm;p;1 (s) of the complex variable s which is a sub-series involving terms over only the tower
(m, p, 1) whose 1st term is a Class II integer as
∞
T
X λ(mpr )
Fm;p;1 (s) = (3.14)
ms prs
r=1

Page 162
8 K. Eswaran

With the understanding that when u = 1 we use the function in (3.14) instead of (3.13), we may write F (s)
as XXX
T
F (s) = Fm;p;u (s). (3.15)
m p u

Comparing the above Eq.(3.15) with Eq(3.10) one can easily see that each term which appears as a
T
summation in (3.10) is actually a sub-series over some tower which we denote as Fm;p;u (s) in (3.15). So we
see that F (s) has been broken up into a number of sub-series. The important point to note is that the λ value of
each term in the sub-series changes its sign from +1 to -1 and then back to +1 and -1 alternatively. Therefore if
the starting value of λ at the base was +1 then the cumulative contribution of this tower (sub series) to L(N )
as N, the upper bound, increases from N to N + 1, N + 2, N + 3, .... will fluctuate between be 0 and 1. For
some other tower whose base value of λ is −1 its cumulative contribution to L(N) will fluctuate between 0 and
−1; these cumulative contributions can be represented in the form of a rectangular wave as shown in Figure 1.
We have arrived at a critical point in our paper. We have cast the original function F (s) ≡ ζ(2s)/ζ(s) as a
sum of functions of s. Since the triad (m, pk , u) uniquely characterises all integers, the summations over m, p, k
and u above are equivalent to a summation over all positive integers n, as in (1.2), though not in the order
n = 1, 2, 3, 4, .... The manner in which the triads were defined ensures that there are neither any missing integers
nor integers that are duplicated.(See Theorems A and B in Appendix II.)
Although we did not explicitly do it, we mentioned in passing that the sum over k in (3.13)and (3.14) is
readily performed since it is a geometric series (see (3.11)) that rapidly converges. This is true not merely for
Re(s) > 1 but also as Re(s) → 0. Whether F (s) converges when the summation is carried out over all the towers
(m, p, u) and, if so, over what domain of s is the central question that we seek to answer in the next section. The
answer to which as we shall see determines the analyticity of F(s) and thus resolves the Riemannian Hypothesis.
We can recast (3.15), still in the domain Re(s) > 1, in the form
∞
X h(n)
F (s) = , (3.16)
ns
n=1

where h(n) is a function appropriately defined below.

By construction, every n in the above summation can be written as

n = µπ ρ ν, (3.17)

where µ, π, and ρ are positive integers, π is the largest prime in the factorization of n, with either (i ) an exponent
ρ ≥ 2, and ν is the product of primes larger than π but with exponents equal to 1 (for Class I integers) or (ii )
it is the largest prime factor with ρ = 1 and ν = 1 (for Class II integers).
We define h(n) as follows:

h(n) = λ(mpk u) if µ = m and π = p and ν = u 6= 1 and ρ = k > 1 (3.18a)

h(n) = λ(mpk ) if ν = u = 1 and π = p and ρ = k ≥ 1 (3.18b)
h(1) = 1 by definition. (3.18c)

Note for all n > 1, (3.18a) and (3.18b) taken together, defines the h(n) for all Class I and Class II integers n.
The factors ms pks us and ms pks in the denominators of (3.13) and (3.14) are simply ns , where n is the integer
characterized by the (m, pk , u) triad (with u = 1 in the latter case).

4 Representation of the summatory Liouville function L(N)

We are now in a position to examine the summatory Liouville function L(N ) and to depict the sum for any
given finite N, as arising from individual contributions from ’rectangular waves’.
To do all this systematically, we will explicitly illustrate the process starting from N = 1, 2, 3... up to
N = 15. Each of these numbers is factored and expressed uniquely as a triad. The N=1 is a constant term,
which is the trivial (1, 1, 1), then the next number N = 2 = (1, 2, 1), is contained in the tower shown below
the one corresponding to N = 1; and N = 3 = (1, 3, 1), is the tower below the previous; 4 = (1, 22 , 1) however
4 is already contained in the tower (1, 2, 1) as its second member; the next N’s: 5, 6, 7, give rise to the new
towers (1, 5, 1), (2, 3, 1), (1, 7, 1); 8 of course is the third member of the old tower (1, 2, 1) similarly 9 is the 2nd
member of (1, 3, 1). After this the new towers which make their appearance are: 10 = (2, 5, 1), 11 = (1, 11, 1), 13 =
(1, 13, 1), 14 = (2, 7, 1) and 15 = (3, 5, 1). Figure 1 shows these and numbers up to N=30.

Page 163
 A Proof of RH 9

Fig. 1. The cumulative sum, L(N) (see top), is obtained by ‘filling up’ slots in various towers from the bottom
up until we have exhausted all N integers.

Now each tower (m, p, u) contributes to L(N ) (consider N fixed in the following) according to the following
rules:
(i) A particular tower will contribute only if its base number is less than or equal to N, i.e. m.p.u ≤ N
(ii) And the contribution C to L(N ) from this particular tower will be exactly as follows:
Case A; Class II integer (u = 1)
C = ΣR r R
r=1 λ(m.p .1) , where R is the largest integer such that m.p ≤ N
Case B; Class I integer (u > 1)
C = ΣK k K
k=2 λ(m.p .u), Where K is the largest integer such that m.p .u ≤ N
Now since each successive λ changes sign from +1 to −1 or vice a versa, the contributions of each tower
can be thought of as a rectangular wave of ever-increasing width but constant amplitude −1 or +1, see Figure
1.
To find the value of L(N), (N fixed), all we need to do is count the jumps of each wave: as we move from
N=0 a jump upwards is called a positive peak, a jump downwards is a negative peak. Draw a vertical line at
N, we are assured that it will hit one and only one peak (positive or negative) in one of the sub-series; then
count the total number of positive peaks P(N) and negative peaks Q(N), of the waves on and to the left of this
vertical line, then L(N ) = P (N ) − Q(N ); the reason for this rule will be clear after the next section.
For an example, take N = 5. There is a positive peak for the constant term (1,1,1), the next wave (1,2,1)
contributes one negative peak (at 2) and a positive peak (at 4), the wave (1,3,1) contributes a −1 peak (at 3)
and (5,1) contributes a −1 peak (at 5). Thus a total of three negative peaks and two positive peaks add up to
give L(5) = −1, which is of course correct. Now if we take N = 10, and draw a vertical line at N=10, looking at
this line and to its left we see that there are additionally three positive and two negative peaks thus adding this
contribution of +1 to the previously calculated value L(5) we get L(10) = 0. (Two red vertical lines just just
beyond N=5 and N=10 are drawn for convenience.) Now if we wish to compute L(15) we see that there are three

Page 164
10 K. Eswaran

more negative peaks and two positive peaks thus giving a value L(15) = −1. Counting the peaks further on it
is easy to check that L(N ) is correctly predicted for every value of N up to 30 and in particular, L(20) = −4,
L(26) = 0 and L(30) = −4.
In summary, to calculate L(N) we merely need to count the negative and positive peaks of the waves on N
and to the left of N. In the figure we have drawn a number of waves and labeled the tower to which each belongs
using a triad of numbers. They are sufficient for one to easily calculate L(N) up to N=30 and check them out
by comparing the numbers with the plot of L(N) shown on the top of the figure.
We turn to a more fundamental point: We show, in Section 6, that, for sufficiently large N (see Appendix
IV), the distribution of the value of L(N) is equivalent to that obtained from summing the distribution of N
coin tosses.

5 Determination of Analyticity of F (s) using Littlewood’s Theorem

We now utilize a technique introduced by Littlewood (1912), to examine the analyticity of the function F (s).
In this, we follow the treatment of Edwards (1974, pp. 260-261).
We have seen that there are two equivalent expressions F (s) viz. Eqs.(1.2) and (3.10) both of which are
given in the form of a series and are absolutely convergent in the region Re(s) > 1. We will therefore follow the
following two procedures:
(i) By using Littlewoods technique we will analytically continue Eq (1.2), which is convergent for Re(s) > 1
to regions Re(s) < 1 and then see that his theorem determines a condition on L(N) for N large, for RH to be
true.
(ii) Similarly instead of using Eq (1.2) we use the equivalent (3.10) and use Littlewood’s technique to
analytically continue Eq (3.10) which is convergent for Re(s) ¿ 1 to regions Re(s) ¡ 1 . This also gives a same
condition as (a) on L(N) for N large for RH to be true. But this time the condition can be interpreted by
a FIGURE. And the FIGURE reveals a clear analogy with coin tosses. Strictly speaking our treatment (ii)
is redundant except for the understanding of the connection of L(N) with coin tosses. In fact for the rest of
the paper we do not need Eq. (3.10) or the Figure, except that the concept of Towers and the factorizations
of integers and the determination of their membership to different towers would be needed to prove several
theorems.

5.1 Littlewood’s theorem applied to F(s) viz. Eqs.(1.2) & (3.10)

We have seen that (1.2) we define:
∞
X λ(n)
F (s) = , (5.19)
ns
n=1

Similarly the alternative expression (3.10) written in the form Eq. (3.16) is:
∞
X h(n)
F (s) = , (5.20)
ns
n=1

with the definition given in (3.18).

Since both of the above expressions are similar in form we use the following generic expression for the
purpose of analysis:
∞
X g(n)
F (s) = , (5.21)
ns
n=1

where g(n) can mean λ(n) or h(n) as the case may be.
The above series (5.19) can be expressed as the integral

Z∞
F (s) = x−s dG(x), (Re(s) > 1), (5.22)
0

Rx
where G(x) = 0 dG is a step function that is zero at x = 0 and is constant except at the positive integers, with a
jump of g(n) at n. The value of G(n) at the discontinuity, at an integer n, is defined as (1/2)[G(n − ) + G(n + )],

Page 165
 A Proof of RH 11

Pn−1
which is equal to j=1 g(j) + (1/2)g(n). Assuming Re(s) > 1, integration by parts yields

Z∞ Z∞
−s
F (s) = d[x G(x)] − G(x)d[x−s ] (5.23)
0 0
ZX
−s
= lim [X G(X) + s G(x)x−s−1 dx
X→∞
0
Z∞
= s G(x)x−s−1 dx, (5.24)
0

where the last step follows from the fact that | G(X) |≤ X, which implies that X −s G(X) → 0 as X → ∞. We
further observe, following Littlewood (1912), that as long as G(X) grows less rapidly than X a for some a > 0,
the integrals in (5.21) and in the line preceding it converge for all s in the half-plane Re(a − s) < 0, that is, for
Re(s) > a. By analytic continuation, F (s) converges in this half-plane. Since this result will be important in
what follows, we record it here.

Theorem [Littlewood (1912)]: When G(X) grows less rapidly with X than X a for some a > 0, F (s) is
analytic in the half-plane Re(s) > a.
We have obtained the above generic result for the analytic continuation of the function given by Eq(5.21).
We will now apply it for the case (i) i.e. Eq. (5.19),in this case g(n) ≡ λ(n) and X ≡ N thus making
G(X) ≡ L(N ). Thus the above Littlewood’s Theorem becomes the condition for the analytic continuation
of(5.19) and which is now restated to read:

Theorem 1 [Littlewood (1912)]: When L(N ) grows less rapidly with N than N a for some a > 0, F (s) is
analytic in the half-plane Re(s) > a.
We shall now demonstrate that the sufficient condition stated in Theorem 1 is satisfied for a specific value
of a that settles the Riemann Hypothesis. (It will turn out that a = 1/2).
Now before devoting the rest of the paper to show that the above condition holds for RH. We will use the
analysis for the analytical continuation of (5.20) i.e. Case (ii).
Hence our definition of G(N ) becomes
N
X
G(N ) = g(n), (5.25)
n=1

and we may rewrite G(N ) as

XXXX
(1 − δu,1 ).(1 − δk,1 ) λ(mpk u) + δu,1 λ(mpk ) ,

G(N ) = (5.26)
m p u k

where δu,1 and δk,1 are Kronecker deltas (e.g. δu,1 = 1 if u = 1 and 0 otherwise). The summations over m, p,
k, and u in (5.26) are undertaken with the understanding that the triads (m, pk , u) will only include integers
n ≤ N . Since the summation over k is over an individual tower(if we keep (m,p,u) fixed we can write(5.26) as
XXX
T
G(N ) = Fm,p,u (s = 0), (5.26b)
m p u

T
This is nothing but Eq.(3.15) evaluated from each subseries Fm,p,u (s) by making s → 0.
Of course, what we have called G(N ) is really the summatory Liouville function, L(N ), defined earlier by
(1.2b), because each integer n occurs only once in the r.h.s. of (5.26b) as an argument of λ i.e. λ(n), Therefore
the G(N ) is really L(N ), hence
N
X
L(N ) = λ(n). (5.24)
n=1

From now on, we revert to the original definitions of the sequence h(n) ≡ λ(n) and G(N ) ≡ L(N ) as defined
in Eq. (1.2) but we may write them in the forms derived in Section 2 using triads.

Page 166
12 K. Eswaran

5.2 Derivation of Theorems concerning Factorization Sequence of λ0 s and the Final Proof of the
Riemann Hypothesis
In this subsection we will derive several crucial theorems concerning the sequence of λ0 s.
Expression (5.26) is crucial because, in the light of Theorem 1, its behaviour will determine the validity of
the Riemann Hypothesis. Every term in the summation in (5.26) is either +1 or −1. We need to determine, for
given N , how many terms contribute +1 and how many −1, and then determine how the sum G(N ) varies with
N.
As we go through the list n = 1, 2, 3, · · · , N , we are assigning the integers to various sets of the kind Pm;p;u .
To use our terminology of towers, we shall be ‘filling up’ slots in various towers from the bottom up until we have
exhausted all N integers. (When N increases, in general, we shall not only be filling up more slots in existing
towers but also adding new towers that were previously not included.) So the behaviour of G(N ) is determined
by how many of the numbers that do not exceed N contribute +1 and how many −1.
It is convenient to identify the λ of an integer by the triad which uniquely defines that integer. To avoid
abuse of notation, we shall denote the value λ(n) in terms of the λ-value of the base integer of the tower to
which n belongs. We will define the λ of the base of a tower in uppercase, as Λ(m, p, u). In other words if
n = (m, pρ , u) then it will belong to a tower whose base number is nB ≡ (m, pκ , u), where κ = 2 if u 6= 1 and
κ = 1 if u = 1. Now we define Λ(m, p, u) = λ(nB ) = λ(mpκ u) = λ(m)λ(pκ )λ(u), since the λ of a product of
integers is the product of the λ of the individual integers. Of course, once we know λ(nB ) we will know the λ of
all other numbers belonging to the tower because they alternate in sign.
To determine the behaviour of G(N ), the following theorem is important.
Theorem 2: For every integer that is the base integer of a tower labeled by the triad (m, p, u), and therefore
belonging to the set Pm;p;u , there is another unique tower labeled by the triad (m0 , p, u) and therefore belonging
to the set Pm0 ,p,u with a base integer for which Λ(m0 , p, u) = −Λ(m, p, u).
Proof:
Let us write the integers at the base of a tower in the form n = mpρ u described by the triad (m, p, u),
where we shall assume that ρ = 2 if u 6= 1 and ρ = 1 if u = 1. These correspond to the smallest members of sets
of Class I and Class II integers, respectively, which are the integers of concern here. In the constructions below,
we shall multiply (or divide) m by the integer 2. Since 2 is the lowest prime number, such a procedure does not
affect either the value of p or u in an integer and so we can hold these fixed.
We begin by excluding, for now, triads of the form (1, p, 1), integers which are single prime numbers. We
allow for this in Case 3 below.
Case 1: Suppose m is odd. We choose m0 = 2m, then
Λ(m0 , p, u) = Λ(2m, p, u) = −Λ(m, p, u). We may say that (m, p, u) and (m0 , p, u) are ‘twin’ pairs in the
sense that their Λs are of opposite sign. Note that (m, p, u) and (m0 , p, u) are integers at the base of two different
towers; they are not members of the same tower. (Recall that the members of a given tower are constructed by
repeated multiplication with p.)
Case 2: Suppose m is even. In this case, we need to ascertain the highest power of 2 that divides m. If m is
divisible by 2 but not by 22 , assign m0 = m/2. (So m = 6 gets assigned to m0 = 3, and m = 3, by Case 1 above,
gets assigned to m0 = 6.) More generally, suppose the even m is divisible by 2k but not by 2k+1 , where k is an
integer. Then, if k is even, assign m0 = 2m; and if k is odd, assign m0 = m/2. (So m = 12 = 22 × 3 gets assigned
to m0 = 23 × 3 = 24. And, in reverse, m = 24 = 23 × 3 gets assigned to m0 = 24/2 = 12.)
Thus for odd m the following sequence of pairs (twins) hold:
(m, p, u) and (2m, p, u) are twins at bases of different towers having λs of opposite signs,(this is Case 1),
(22 m, p, u) and (23 m, p, u) are twins at bases of different towers having λs of opposite signs,
(24 m, p, u) and (25 m, p, u) are twins at bases of different towers having λs of opposite signs,
and so on.
Case 3: Now consider the case where the triad describes a prime number; that is, it takes the form (1, p, 1).
For the moment, suppose this prime number is not 2. In this case, where m = u = 1, we simply assign m0 = 2.
Clearly,
Λ(2, p, 1) = −Λ(1, p, 1), and the numbers (2, p, 1) and (1, p, 1) are at the bases of different towers.
Case 4: Finally, consider the case where the triad describes a prime number and the prime number is 2;
that is, the integer (1, 2, 1), for which Λ(1, 2, 1) = −1. We match this prime to the integer 1. By definition
λ(1) = Λ(1, 1, 1) = 1. Thus the first two integers have opposite signs for their values of λ. 
So, in partitioning the entire set of positive integers, the number of towers that begin with integers for
which λ = −1 is exactly equal to those that begin with integers for which λ = +1.

Thus, Theorem 2 immediately gives the following result:

Page 167
 A Proof of RH 13

Theorem 3: In the set of all positive integers, for every integer which has an even number of primes in its
factorization there is another unique integer, (its twin), which has an odd number of primes in its factorization.

The consequence of the above theorem follows not just from that each integer has a unique twin whose
λ-value is of the opposite sign, but also from from the context that these lie in an alternating sequence.
That is, not only are the bases of two uniquely-paired towers twins, the next higher number in the first
tower is the twin of the next higher number in the second tower, and so on. Thus every integer with
λ = ±1 in a unique tower is twinned uniquely in alternating sequence with the integers with λ = ∓1 in
another unique tower, ensuring that the proportions of numbers with λ = +1 and λ = −1 are equal over
the entire natural number system. Alternately, it may be argued that each Tower, which is an ordered
infinite sub-set of the natural number system, is itself equally partitioned into members with λ = +1 and
λ = −1 by their alternating sequence in that order. As the natural number system (excluding 1) comprises
only such Towers, it too is so equally partitioned, and has equal proportions of numbers with λ = +1 and λ = −1.

Thus we have shown:∗

Theorem 3B : If n is an arbitrary positive integer,

P rob[λ(n) = +1] = P rob[λ(n) = −1] = 1/2. (19)

The result in Theorem 3B is a necessary condition for RH to be valid; it is not sufficient.† The condition
that is equivalent to proving RH ((Littlewood (1912), Edwards (1974)) is the following:

L(N )
1 = 0, as N → ∞, (20)
N 2 +
for any  > 0. We describe how this equivalence is formally proved below.

1. We compare the L(N ) series to X(N ), the distance travelled in N unit steps in a standard random walk,
which can be represented as:
XN
X(N ) = c(n) (21)
n=1

where the c(n)’s are independent random numbers with an equal probability of being either +1 or -1, i.e.,
“coin-tosses”. It is a well-known result (see Chandrasekhar (1943)) that the expected value of |X(N )|, for
large N , is
limN →∞ E(|X(N )|) = C0 N 1/2 (22)
The further line of advance of the proof is to show now that Equation (22) applies to L(N ) as well, and
so proves Equation (20), and thereby the RH. To do this we have to prove that the λ(n)’s in the L(N )
series are essentially “coin-tosses”, for large n.

2. To show that the λ’s behave as coin tosses, we have to show that (i) their probabilities of being either +1
of -1 are equal, as was proved by Theorem 3B, further (ii) we have to show that the λ’s appearing in the
natural sequence, n = 1, 2, 3, ..., are independent of each other — i.e., that the value of λ(n) has no influence
on the value of λ(n + 1), say. This seems counter-intuitive, as the λ’s are obviously deterministically linked.
Nevertheless, their independence in the natural sequence is shown by two different approaches:
(a) In Appendix III, it is proved that the sequence of λ(n), n = 1, 2, 3, ... is non-cyclic. This would preclude
any dependence of the type
λn = f (λn−1 , λn−2 , λn−3 , ..., λn−M )
because any finite series of +1’s and -1’s of length M would have a finite number of permutations
P , so the series λn−1 , λn−2 , λn−3 , ..., λn−M must repeat itself after atmost P numbers, and thereafter
become cyclic if such a dependence relationship exists between the λ’s. The non-cyclic nature of L(N )
conforms also to the notion of randomness in Knuth (1968, Ch. 3).

∗ These Theorems 3 and 3B, were later given new, alternate proofs, which only need induction and the starting premise that every
odd integer starting from 1, has a unique successor integer, which is even and with which it forms a unique ‘Pair’ and that the even
integer in every such Pair, has an odd integer which is its predecessor and its ‘partner’, the successor of an even integer is an odd
integer not its ‘partner’. See: K. Eswaran, (April 2018)
† Borwein et al (2006, p. 48) claim that the result in Theorem 3 is equivalent to a proof of RH.

Page 168
14 K. Eswaran

(b) In Appendix IV, another approach is taken. It is shown that from merely these two rules: λ(p) = −1 for
a prime p, and λ(pq) = λ(p) × λ(q), where q is any integer, the entire sequence of λ’s for n = 1, 2, 3, ...,
can be obtained without determining the number of prime factors of n. It is then argued that for
any two integers n > m, λ(n) is dependent on λ(m), if the latter is required to find the former,
and independent if not. It is then shown that, as n → ∞, any finite sequential strip of λ’s will be
independent of each other, thus essentially making them equivalent to coin-tosses.
3. With Theorem 3B and Appendices III and IV, we have proved above that the L(N ) series is one realization
of a random walk. We then invoke (towards the end of Section 5 in [5]) Khinchin and Kolmogorov’s law of
the iterated logarithm to show that the maximum deviation dN , from the one-half power-law expectation,
in the exponent of |X(N )| for any individual random walk tends monotonically to zero as N → ∞. So (22)
also holds for |L(N )|. Therefore for any chosen  > 0 in (20), the statement for the validity of RH will be
satisfied and the Riemann Hypothesis is proved.
4. In Appendix V, starting from Littlewood’s ansatz, that L(N ) ∼ N a , for N → ∞, we argue that the
statistics of the λ’s must become “self-similar”, i.e., independent of N for large N . This principle yields
us the value of a = 21 , which again would satisfy (20). This ‘physicist’s proof’ of the RH, is separate from
the argument in the main paper, and may be treated as an interesting addendum to it.
5. Finally, in Appendix VI, we confirm by considering the λ’s from n = 1 to 176 trillion, that their sequence is
statistically indistinguishable (using the χ2 statistical test) from coin-tosses over the entire set of numbers
considered, and also when it is partitioned in smaller sections. While this is merely a ‘verification’, not a
‘proof’, this fact has not been reported in literature, and by itself, requires an explanation (which we have
provided) given its surprising nature.

With Theorem 3B and Appendices III and IV, we have proved that the L(N ) series is a random walk. We
formally confirm this below.
N
P
Theorem 4: The summatory Liouville function, L(N ) = λ(n) has the following asympotitic behaviour:
n=1
1
|L(N )| ≤ C0 N 2 +dN as N → ∞.

Proof: Theorem 3B gives P r(λ(n) = +1) = P r(λ(n) = −1) = 1/2 , where P r denotes probability. Given the
results in Appendicies III and IV, the λ-values behave like ‘ideal coin’ tosses, where λ(n) = +1 as head and
λ(n) = −1 as tail, and L(N ) is the cumulative result of N successive coin tosses, and is equivalent to the
distance X(N ) moved in a random-walk with N unit steps. Chandrasekhar (1943) has shown that, for a random
1
walk of N steps, Expectation(|X(N )|) = C0 N 2 as N → ∞. The quantity dN (≥ 0) seen above is the maximum
deviation from expectation of an individual random walk of N steps. QED

We conclude this section by estimating the ‘width’ of the Critical Line, the region around Re(s) = 1/2
in which the non-trivial zeros of ζ(s) must lie. Invoking Littlewood’s Theorem (Sec.5), we deduce that
F (s) ≡ ζ(2s)/ζ(s) is analytic in the region a = 1/2 + d∞ < s < 1 (where d∞ ≡ limN →∞ dN ). This implies
ζ(s) has no zeros in the same region. But Riemann had shown by using symmetry arguments‡ that if ζ(s)
has no zeros in the latter region then it will have no zeros in the region 0 < s < 1/2 − d∞ ; taking both these
results together we are lead to the conclusion that all the zeros can only lie in the 1/2 − d∞ < Re(s) < 1/2 + d∞ .

It is interesting that the law of the iterated logarithm enunciated by Kolmogorov (1929), also see
Khinchine (1924), gives an expression for dN . The statement of the law, adapted to the present context, is:
Let {λn } be independent, identically distributed random variables with means zero and unit variances. Let
SN = λ1 + λ2 + . . . + λN . The limit superior (upper bound) of SN almost surely (a.s.) satisfies

SN
Lim Sup √ =1 as N → ∞
2N log log N

Now, from Theorem 4 we have written that if we consider the λ0 s as “coin tosses” one can write
1
L(N ) = λ1 + λ2 + . . . + λN ≤ C0 N 2 +dN (as N → ∞) (since we are interested in only the behaviour for
large N we henceforth ignore the constants). Comparing this expression with the one above we see that one

‡ He did this first by defining an associated xi function: ξ(s) ≡ Γ(s/2)π s/2 ζ(s), Γ(s) is the Euler Gamma function, then showed that
this xi function has the symmetry property ξ(s) = ξ(1 − s) which in turn implied that that the zeros of ζ(s) (if any) which are not on
the critical line will be symmetrically placed about the point s=1/2, ie.if ζ( 21 + u + iσ) is a zero then ζ( 12 − u − iσ), (0 < u < 1/2),
is a zero see Whittaker and Watson page 269.

Page 169
 A Proof of RH 15

1 √
can write N 2 +dN ∼ N log log N thus yielding an expression§ for dN = log2log log N
log N . We see that dN → 0
as N → ∞. So the equivalent statement Equation (20) will be satisfied for any chosen  > 0. Further
d∞ (≡ limN →∞ dN ) is the half-width of the critical line. Since this is zero, we conclude that all the non-
trivial zeros of the zeta function must lie strictly on the critical line. Thereby, the Riemann Hypothesis is proved.

6 Conclusions
In this paper we have investigated the analyticity of the Dirichlet series of the Liouville function by constructing
a novel way to sum the series. The method consists in splitting the original series into an infinite sum over
sub-series, each of which is convergent. It so turns out each sub-series is a rectangular function of unit amplitude
but ever increasing periodicity and each along with its harmonics is associated with a prime number and all
of them contribute to the summatory Liouville function and to the Zeta function. A number of arithmetical
properties of numbers played a role in the proof of our main theorem, these were: the fact that each number can
be uniquely factorized and then placed in an exclusive subset, where it and its other members form an increasing
sequence and factorize alternately into odd and even factors and thus have equal proportions of numbers with
λ = +1 and -1; and each subset can be labelled uniquely using a triad of integers which in their turn can be
used to determine all the integers which belong to the subset. This helped us to show that for every integer that
has an even number of primes as factors (multiplicity included), there is an integer that has an odd number of
primes. This provides a proof for the long-suspected (Denjoy 1931) but unproved conjecture—until now—that
the Riemann Hypothesis has a connection with the coin-tossing problem. Further, it has now been revealed
that the randomness of the λ(n)’s in the natural sequence¶ is the reason that the non-trivial zeros of the Zeta
function all lie on the critical line: Re(s) = 1/2.

7 DEDICATION
I dedicate this paper to my teachers: Mr John William Wright of Bishop’s School Poona, Prof. S.C. Mookerjee
of St. Aloysius’ College Jabalpur, Prof. P.M.Mathews of Univerity of Madras, Mr. D.S.M. Vishnu of BHEL R&
D Hyderabad and to my first teachers - my parents. All of them lived selfless lives and nearly all are now long
gone: May they live in evermore.

Acknowledgments: I thank my wife Suhasini for her unwavering faith and encouragement. I thank my
brothers Mukesh and Vinayak Eswaran, Dr. S.V Ramanan and Prof. George Reuben Thomas for carefully going
over the manuscript and for their suggestions.
I acknowledge with grateful thanks my email correspondence with Prof. Sir Michael Berry, Professor of
Physics (emeritus) Univ. of Bristol and Prof. German Sierra, Inst. of Theoretical Physics, UAM-CSIC, Madrid,
which has helped explain the various intricate arguments expounded in the Extended Abstract of this paper.
I also thank the support of the management of SNIST, viz. Dr. P. Narsimha Reddy and Dr. K.T. Mahi, for
their constant support, and my departmental colleagues and the HOD Dr. Aruna Varanasi for providing a very
congenial environment for my research.I sincerely thank Mr.Abel Nazareth of Wolfram Research for providing
me a version of Mathematica which made some of the reported calculations in Appendix VI, possible.

8 APPENDIX I: Scheme of partitioning numbers into sets

Our scheme of partitioning numbers into sets is as follows:

(a) Scheme for Class I integers:
Let us say n = pe1 pf2 pg3 ...phm pkL pj pt , then it will have at least one prime which has an exponent of 2 or above
and among these there will a largest prime pL whose exponent is atleast 2 or above. Such a prime will always
exist for a Class I number. Then by definition the number to the right of pL is either 1 or is a product of primes
with exponents only 1. Now multiply all the numbers to the left of pL and call it m i.e. m = pe1 pf2 pg3 ...phm and the

§N 1
√ √ 1 1 1 1 1
2
+dN
∼ N log log N = elog N log log N = e 2 log{N log log N } = e 2 log N + 2 log log log N = N 2 e 2 log log log N which then implies
1 log log log N
e2 = N dN = edN log N thus giving dN = log2log log N
log N
¶ God seems to have “played dice” at least once, when he created the natural number system!

Page 170
16 K. Eswaran

product of numbers to the right of pL as u i.e. u = pj pt .Now this triad of numbers m, pL , u will be used to label
a set,note n = m.pkL .u Let us define the set Pm;pL ;u :

Pm;pL ;u = {m.p2L .u, m.p3L .u, m.p4L .u, m.p5L .u, m.p6L .u, m.p7L .u, ....} (A1)

Obviously n = m.pkL .u which has k ≥ 2 belongs to the above set. Also notice the factor involved in each number
increases by a single factor of pL therefore the λvalues of each member alternate in sign:

λ(m.p2L .u) = −λ(m.p3L .u) = λ(m.p4L .u) = −λ(m.p5L .u) = λ(m.p6L .u) = .......(A2)

In this paper ALL sets defined as Pm;p;u will have the property of alternating signs of λ Eq. (A1). Note in the
above set containing only Class I integers m will have only prime factors which are each less than pL .
Let us consider various integers:
Ex 1. Let us consider the integer 73573500; this is factorized as 22 .3.53 .73 .11.13 and since this is a Class I
integer, and pL = 7 because 7 is the highest prime factor whose exponent is greater than one. pL = 7 m = 22 .3.53
and u = 11.13 = 143 and therefore 73573500 is a member of the set P1500;7;143

P1500;7;143 = {1500.72 .143, 1500.73 .143, 1500.74 .143, 1500.75 .143, ...}

Ex 2. Now let us consider the simple integer: 34 this is a class I integer and belongs to P1;3;1 =
{3, 32 , 33 , 34 , 35 , 36 , ......}

Ex 3. Let us consider the integer 663 this is factorized as: 3.13.17 and is a Class II integer as there no
exponents greater than 1, and 663 = 3.13.17 and since 17 is the highest prime number we put this in the set:

P39;17;1 = {39.17, 39.172 , 39.173 , 39.174 , ....}.

NOTE: If a tower has a Class II integer then it will appear as the first (base) member, all other numbers will
be Class I numbers.
Ex 4. Let the integer be the simple prime number 19, we write:

19  P1;19;1 = {19, 192 , 193 , 194 , .....}

Ex 5. Let the integer be 4845 this is factorized as3.5.17.19 since this is a Class II integer we see
m = 3.5.17 = 255, p = 19, u = 1 and the set which it belongs is

P255;19;1 = {255.19, 255.192 , 255.193 , 255.194 , 255.195 , ....}

9 APPENDIX II: Theorems on representation of integers and their partitioning into

sets.
Theorem A: Two different integers cannot have the same triad (m, pk , u)
Let a and b be two integers which when factored according to our convention are a = n.q g .v and b = n0 .q 0h .v 0 ,
and let us consider only Class I integers u, v and v 0 are all > 1.
If they are both equal to the same triad (say) (m, pk .u). Then m.pk .u = n.q g .v = n0 .q 0h .v 0 . Consider the
first two equalities m.pk .u = n.q g .v, which means p is the largest prime with k > 1 on the l.h.s. Similarly q is
the largest prime with exponent g > 1 on the r.h.s. Now if p > q this means pk must divide v, but this cannot
happen since v cannot contain a prime greater than q with an exponent k > 1. Now if p < q then q g must divide
u but this again cannot happen since u cannot contain an exponent g > 1. So we see p = q, and k = g. But once
again unique factorization would imply, since u contains all prime factors larger than p and v must contain only
prime factors larger than q(= p), the only possibility is u = v, but this also makes m = n. That is, the triad of
a is (m, pk , u). Similarly equating the second and third equalities n.q g .v = n0 .q 0h .v 0 and using similar arguments
we see n = n0 , q = q 0 , and v = v 0 ; that is, a = b. The same logic can be used to prove the theorem for class II
integers when u = v = v 0 = 1. QED.

Theorem B: Two different triads cannot represent the same integer.

If there are two triads (m, pe , u) and (m0 , rs , u0 ) and represent the same integer say a which can be factorized
as a = n.q g .v. Where the factorization is done as per our rules then we must have m.pe .u = n.q g .v by using
exactly similar arguments as above(in Theorem A) we conclude that we must have m = n, p = q, e = g and
u = v; similarly imposing the condition on the second triad m0 .rs .u0 = n.q g .v, we conclude m0 = n, r = q, s = g
and u0 = v; thus obtaining m = m0 , p = r, e = s and u = u0 this means the two triads are actually identical. QED

Page 171
 A Proof of RH 17

10 APPENDIX III: Non-cyclic nature of the factorization sequence

It is a necessary condition in the tosses of an ideal coin that the results are not cyclic asymptotically, namely
the results cannot form repeating cycles as the number of tosses becomes large.
Definition
Let nk be the number of primes, repetitions counted, in the factorization of a positive integer k. We call
{n1 , n2 , ..., nk , ....} the factorization sequence.
Note: λ(k) = +1 if nk is even and λ(k) = −1 if nk is odd.
Theorem The factorization sequence is asymptotically non-cyclic.
Proof: The result follows from this claim:
Claim. The sequence λ(1), λ(2), λ(3), ..., λ(n), ..., is asymptotically non-cyclic.
If the claim is not true there would exist an integer t, t ≥ 0, so that the sequence is cyclic (after λ(t)), with
cycle length σ.
By Theorem 3, the number of positive integers with even number of prime factors (counting multiplicities)
equals the number of positive integers with odd number of prime factors (counting multiplicities). Therefore,
the λ’s in each cycle must sum to zero as do the first t λ’s before the cycles start.
Then L(N ) ≤ max{t/2, σ/2}.
Now we use Littlewood’s Theorem 1 and noting that in (5.21) G(x) ≡ L(x), we substitute the maximum
value of L(x) as x → ∞, viz. | L(x) |= σ/2, and thus deduce that (5.21) will always converge provided 0 < s.
Since, | L(x) |≤ σ/2, L(x) indeed grows less rapidly than xa for all a > 0, satisfying the condition in Theorem
1. This means that we should be able to analytically continue F (s) ∼ ζ(2s)/ζ(s) leftwards from Re(s) = 1 to
Re(s) = 0, contradicting Hardy (1914) [3] that there are very many zeros at Re(s) = 1/2 and these will appear
as poles in F (s). This proves the Claim.QED.

11 APPENDIX IV: The sequence of λ’s in L(N), are equivalent to Coin Tosses
In this paper we showed in Theorem 3, that the λ(n) have an exactly equal probability of being +1 or −1. Then
in Appendix III, we showed that the sequence λ(1), λ(2), λ(3), ..., λ(n), ... can never be cyclic. The latter result
in the minds of most computer scientists would be interpreted as that the sequence of λ’s by virtue of it being
non-repetitive, is truly random,(Knuth (1968); Press etal (1986)) and hence it is legitimate√to treat the sequence
as a result of coin tosses and thus one can then say that L(N ) = ΣN 1 λ(n), will tend to N thus proving RH,
by using the arguments given at Section 5.
However, this done, there would be some mathematicians who may remain unconvinced, because we have
not strictly proved that the λ’s in the series are independent. The purpose of this Appendix k is to prove that this
is indeed the case. This allows us to demonstrate the λ−sequence has the same properties as, and is statistically
equivalent to, coin tosses, thus placing ourP proof of RH beyond any doubt.
N
We again consider the series L(N ) = n=1 λ(n), which is re-written as:

N
X
L(N ) = Xn (1)
n=1

It has already been proved in this paper that, over the set of all positive integers, the respective probabilities that
an integer n has an odd or even number of prime factors are equal. So, Xn (= λ(n)), can with equal probability,
be either +1 or -1. It will now be shown that the values of Xi and Xj , i 6= j, are independent of each other, as
n → ∞, and so will become the equivalent of ideal-coin tosses.

11.1 The λ values as a deterministic series

We first show that the λ’s in the natural sequence, far from being random, are actually perfectly predictable
and therefore deterministic. That is, knowing the λ’s (and the primes) up to N , we can directly obtain (without
resorting to factorisation) the λ’s (and primes) up to 2N thus:

We obtain integers m in the range N < m ≤ 2N by multiplying the integers n and q in the range 1 < n, q ≤ N ,
such that N < nq ≤ 2N and then using the property λ(q ∗ n) = λ(q) ∗ λ(n) to find λ(m = q ∗ n). However,
not all the numbers in the range N < m ≤ 2N will be covered by such multiplications. That is, there will
be ‘gaps’ in the natural sequence left in the aforesaid multiplications, where no n and q can be found for
some m’s in N < m ≤ 2N . These m’s will identified as prime numbers. The λ of a prime is -1. Thus, by

kI thank my brother Vinayak Eswaran for providing the kernel of the proof given in this section.

Page 172
18 K. Eswaran

knowing the λ’s and the primes up to N , we can predict the λ’s (and primes) up to 2N . This process can
be repeated ad-infinitum to compute the λ’s of the natural sequence up to any N , from just λ(1)=1 and λ(2)=-1.

We emphasize that any other method of evaluating the λ’s, including direct factorisation, must perforce
yield the same sequence as the method above. Therefore, this method offers a complete description of the
determinism embedded in the series.

11.2 Relationships and dependence between λ’s

We note that every integer n has a direct relationship (which we will call a d-relationship) with all numbers
n ∗ p, where p is any prime number. We can define higher-order d-relatives in the following way: the integers
(n, n ∗ p) are in a first-order d-relationship, (n, n ∗ p ∗ q) are in a second-order one, and (n, n ∗ p ∗ q ∗ r) are in
a third-order one, and so on, where p, q, r are primes (not necessarily unequal).
In the deterministic generation of λ’s outlined above, it is clear that their values will be determined through
d-relationships, which would thereby make their respective values dependent on each other. It is evident that
the λs of two d-relatives n and m(> n), are dependent on each other and that λ(m) = (−1)o λ(n), where o is the
order of the relationship.
There is another kind of relationship we must also consider: we can have a c- (or consanguineous) relationship
between two non-d-related integers m and n if they are both d-relatives of a common (‘ancestor’) integer smaller
than either of them. So we can trace back the λ’s along one branch to the common ancestor and trace it up the
other to find the λ of the other integer. It is convenient to take the common ancestor as the largest possible one,
which would be the greatest common factor of the two integers, which we shall call G.
Now we ask the question, when are m and n not related? When they have neither a d-relationship nor a
c-relationship with each other. That is, when they are co-primes: as then neither integer would appear in the
sequence of multiplications that produce the other by the deterministic iterative method. In such a situation,
the λ of neither is dependent on the other, so their mutual λ’s are independent.
Now consider two c-relatives, m and n, which share the greatest common factor G. We can write m = G ∗ P
and n=G ∗ Q, where P and Q are chosen appropriately. As G is the greatest common factor of m and n, it is
clear that P and Q are co-primes. Now we consider the relationship between λ(m) and λ(n) and explore their
relatedness. This turns out to be self-evident: As λ(m) =λ(G) ∗ λ(P ) and λ(n) =λ(G) ∗ λ(Q), and we know
that λ(P ) and λ(Q) are independent of each other, it follows that λ(m) and λ(n) are also independent of each
other.∗∗

11.3 The unpredictability of λ values from a finite-length sequence: d-relatives

We have concluded above that the only λ’s in L(N ) that are dependent are those between d-relatives, where the
smaller integer is a factor of the other. We see that the distance of two such “first-order” relatives, n and n ∗ p,
from each other is n(p − 1) which increases without bound with n. Further all the first-order d-relatives of n
also have relative distances with each other that are at least as great as n (as their respective p’s will differ at
least by 1). Thus the d-relationship between numbers is a web with increasing distances between their first-order
relatives†† . It is also easy to see that the higher-order d-relatives of any integer n will also be at a distance of at
least n from n itself and from each other.
Now we consider if we would be able to predict λ(N + 1) if we know only the λ’s between N − L < n ≤ N ,
where L is some finite number? We would be able to do so only if N + 1 is a d-relative (of any order) of any of
the numbers N − L < n ≤ N . However, for n large enough the d-relatives of N + 1 will be far from it and would
not come in the range of numbers N − L < n ≤ N . So essentially, there is no way of predicting λ(N + 1) from
the range of L λ’s coming before it. This means λ(N + 1) is independent of the range of L λ’s coming before it.
Therefore, the λ’s on all finite lengths are independent of each other, as N → ∞.

11.4 Closure
We have investigated the dependence of λ’s appearing in L(N ) in the natural sequence n = 1, 2, 3, .., . We first
show that the λ’s are in a perfectly deterministic sequence (which is not random in the slightest way, except
in the unpredictable discovery of primes) that allows us to obtain all of them up to any integer N by knowing

∗∗ It may be noticed that m and n belong to different towers. It is worth mentioning that the arguments made here in Appendix
IV, can be couched in the language of towers as we did in Sections 2 and 3.
†† How rapidly the relationship distance increases can be gauged from the fact that the 2r sequence, which has the slowest increases,
nevertheless will have its 100th element placed at around n ≈ 1030 in the natural sequence, and the distance to the 101st element
will also be 1030 !

Page 173
 A Proof of RH 19

only that λ(1) = 1, λ(2) = −1, λ(q ∗ n) = λ(q) ∗ λ(n), and that λ(p) = −1 for any prime p. We then propose
that the λ’s of two integers m and n can be dependent only if the integers are connected through the sequence
of multiplications involved in the deterministic process. If they are not so related, as would happen if they are
co-primes, their λ’s would be independent. We then investigate the only two possible types of relationships and
show that one, the d-relationship, leads to dependencies between numbers that are increasingly distant. The
other, the c-relationship, is shown to give independent λ’s. The result obtained is that that the λ’s in any finite
sequence are independent, as N → ∞. QED

12 APPENDIX V
An Arithmetical Proof for |L(N )| ∼ N 1/2 as N →∞

In this appendix we provide an alternate, but √ this time an arithmetical, proof of the asymptotic behavior
of the summatory Liouville function, viz.|L(N )| = N as N → ∞ However in order to do this we first require
to prove a theorem on the number of distinct prime products in the factorization of a sequence of integers and
their exponents. We will be considering special types of Sets S −√(N ) and S + (N ) which contain a sequence of
consecutive integers, they are defined below; each are of length N , where in this section N will always be a
perfect square. A collection of all such sets will contain as its members all the integers and the intersection of
any two different sets will be null. See Tables 1.1 to 1.4 in Appendix VI. We will be studying the contribution of
such sets to the summatory function L(N ) in order to determine the asymptotic behavior of the latter as N → ∞

Theorem A5 : Consider the sequence S comprising M (N ) consecutive positive √ integers, defined by S − (N ) =

{N − M (N ) + 1, N − M (N ) + 2, N − M (N ) + 3, ........, N }, where M (N ) = N . Then every number in S − (N )
will firstly belong to different towers,∗ and further every number will: (a) differ in its prime factorization from
that of any other number in S − (N ) by at least one distinct prime† OR (b) in their exponents.
The statement of this theorem can be roughly considered as an extremely weak form of Grimm’s
conjecture,(1969) which states that a sequence of k consecutive composite integers will have at least k distinct
primes in their factorization, also see Ramachandra etal.(1975), Grimm’s theorem though not proved, yet, has
been verified for very many subsets, see: S. Laishram and Ram Murty (2006,2012), and Balasubramanian, etal
[2009]. We do not need this very strong version for our arguments.
We first take up the task to prove (a) because it is by far the more common occurrence. In case condition
(a) does not hold in a particular situation then condition (b) is always true, because of the uniqueness of
factorization.
Proof :
Let there be k primes in the sequence S − (N ). Denote the j integers in the sequence that are not primes
by the products pi bi , i = 1, 2..., j, where pi is a prime and, obviously, k + j = M (N ). Denote the subset of
these non-prime integers by√J. There is no loss of generality if we assume the primes pi in the products pi bi ,
i = 1, 2..., j, to be less than N − 1/2 and also the smallest of prime in the product.‡
To prove the theorem, we compare two arbitrary members, pi bi and pj bj , i 6= j, belonging to set J.
Case 1: Suppose pi 6= pj . If bi 6= bj , bi must contain a prime that does not appear in the factorization of
bj (and hence pi bi must be different from pj bj by this prime). For if bi and bj do not√differ by a prime, we must
have bi = bj ≡ b. This means the difference of pi bi − pj bj = (pi − pj )b is larger than N√in absolute value. This
is not possible since the members of the sequence S − (N ) cannot differ by more than N . Therefore bi must
differ from bj by a prime in its factorization.(One may think that it may be plausible that bi = br and bj = bm ,
where r and m are positive integers, in which case pi bi differs from pj bj only in the prime pi . √However, this
eventuality will never arise because then the difference between pi bi and pj bj will be more than N .)
Case 2: Suppose pi = pj ≡ p then bi and bj must differ by a prime factor or their exponents are different.
Because of ‘unique factorization’, if they do not differ by a prime factor it means pi .bi = pj .bj = p.b, unless the
r0 r0 r0
factors of bi and bj are of the form: bi = pr11 .pr22 ...prkk and bj = p11 .p22 ...pkk which implies that is rl = rl0 is not

∗ Two numbers n = m.pα .u and n0 = m.pβ .u, (n < n0 ), of the same tower, cannot both belong to the set S − (N ) because they will
be too far separated to be within the set, as their ratio n0 /n ≥ p ≥ 2
† For example, if two numbers c and d in S are factorized as c = pe1 pe2 and d = pe3 pe4 then at least one of the primes p or p will
1 2 3 4 3 4
be different from p1 or p2 . √
‡ This is readily seen as follows. Since every member of J lies between N − N and N , clearly any composite member, written as
√ p √ √
a product ab, cannot have both integers a and b less than N − 1/2. (We are invoking the fact that (N − N ) = N − 1/2,
√ 1 √ 1
approximately.) Let a be the smaller of the two numbers, and so a < N − 2 and b > N − 2 . If a is a prime number, set p = a.
If a is not a prime number, factorize it and pick the smallest prime p which is one of its prime factors.

Page 174
20 K. Eswaran

true for all rl , l = 1, 2..k , hence in this case the exponents are different (actually this case is very rare. It can
be shown: the case k = 2 cannot occur and therefore if at all this case occurs, k must be greater than 3).
Since pi bi and pj bj are arbitrary members of the set J, it follows that every integer in J must differ from
another integer in J by at least one prime in its factorization or by its exponent, thus making the λ−values of
any two members of the set S − (N ) not dependent on each other . 

The above theorem has profound implications for the λ-values of the numbers in the sequence S − (N ).
If we take the primes to occur randomly (or at least pseudo-randomly), the λ-value of each of these M (N )
integers—although deterministic and strictly determined by the number of primes in its factorization—cannot
be predicted by the λ-value of any other number in the sequence S − (N ). That is, the λ-value of any number
in S − (N ) can be considered to be statistically independent of the λ-value of another member of this sequence,
primarily because they stem from different towers and also because (as we have proved in A5) any two such
numbers differ by at least one prime. Hence the λ-values in the sequence Sλ− ≡ {λ(N − M (N ) + 1), λ(N −
M (N ) + 2), λ(N − M (N ) + 3), ........, λ(N )}, in which each member has a value either +1 or −1, would appear
randomly and be statistically similar. By this we also deduce that two different sequences of λ−values defined
on two different sets (say) S − (N ) and S − (N 0 ) with N 6= N 0 are statistically similar, because they have the same
properties which also means that they can be separately compared with other sequences of coin tosses and the
comparison should yield statistically similar results.
We will use these deductions to obtain the main result of this appendix viz a = 1/2 in the expression
|L(N )| = N a as N → ∞
Although it is not explicitly required for what follows, we note that it is not hard to prove that the
sequence S + (N ) ≡ {N + 1, N + 2, N + 3, ........, N + M (N )} of length M (N ) also behaves similarly. That is,
every member of S + (N ) satisfy condition (a) OR (b) of the above Theorem for S − (N ) stated above.
The proof mimics the one provided above and so is omitted.§ Hence the λ-values in the sequence Sλ+ ≡
{λ(N + 1), λ(N + 2), λ(N + 3), ..., λ(N + M (N ))}, in which each member has a value either +1 or −1, would
also appear randomly and behave statistically similarly.
√
12.1 Arithmetical proof of |L(N )| ∼ N , as N → ∞
We now show that if the summatory Liouville function
N
X
L(N ) = λ(n), (1)
n=1

takes the asymptotic form

|L(N )| = C N a , (2)
where C is a constant, then we must have:
a = 1/2. (3)
Throughout this subsection we will always assume that N is a very large
√ integer.
Consider the sequence of consecutive integers of length M (N ) = N :
√ √ √
SN = {N − N + 1, N − N + 2, N − N + 3, · · · · · · , N } (4)

Each of the M (N ) integers in the sequence SN can be factorized term by term and would differ from another
member in SN by at least one prime or exponent,(as proved in the above theorem).¶
Now since, N is large, all the primes involved may be considered random numbers (or pseudo-random
numbers), therefore as reasoned above, we can conclude that the λ−sequence associated with SN viz.
√ √ √
{λ(N − N + 1), λ(N − N + 2), λ(N − N + 3), · · · · · · , λ(N )} (5)

will take values which are random e.g.

{−1, +1, +1, , −1, +1, · · · , +1} (6)

√ √
where in the above example λ(N − N + 1) = −1, λ(N − N + 2) = +1 etc. Furthermore, since the λ-values
have an equal probability of being equal to +1 or −1 (Theorem 3) and the sequence is non-cyclic (Theorem 11.1,

§ Thisimplies, interestingly, that by choosing N to be consecutive perfect squares, the entire set of positive integers can be envisaged
as a union of mutually exclusive sequences like S − (N ) and S + (N ).
¶ Therefore, in the terminology of Sections 2 and 3, each of them will mostly belong to different Towers.

Page 175
 A Proof of RH 21

in Appendix 3), the above sequence will have the statistical distribution of a sequence of tosses of a coin (Head
= +1,Tail√= −1). But we already know from Chandrasekhar(1943) that if the λ’s behave like coin tosses then
|L(N )| ∼ N , as N → ∞. However, we do not know whether the entire sequence √ of λ’s occurring
√ in Eq.(13.1)
behaves
√ like coin tosses; for any given N , it√is only the subsequence {λ(N − N + 1), λ(N − N + 2), λ(N −
N + 3), · · · · · · , λ(N )} of length M (N ) = N that does behave like coin tosses.
On the other hand if we had a sequence of length N , of real coin tosses (say) c(n), n = 1, 2....N , where
c(n) = ±1, then the cumulative sum, Lc (N ), of the first N of such coin tosses is given by:
N
X
Lc (N ) = c(n). (7)
n=1

Then for N large we do know from Chandrashekar (1943) that

√
|Lc (N )| ∼ N . (8)
√
We can then estimate the contribution P1/2 to Lc (N ) from the last M (N ) = N terms in Eq.(13.7), this
would be:
N
X
P1/2 = c(n)
√
n=N − N +1
√
= Lc (N ) − Lc (N − N) (9)

Now since Eq.(13.7) represents perfect tosses Eq. (13.9) becomes

√ √ 1 1 1
P1/2 = N − (N − N )1/2 = − √ ,
2 8 N

that is,
P1/2 = O(1) (10)
√
In Eq. (13.10), P1/2 is the contribution to Lc (N ) from the last M (N ) = N tosses of a total of N tosses of a
coin. We shall consider the value of P1/2 as the benchmark with which to compare the contributions of the last
√
M (N ) = N terms of the summatory Liouville function.
Now coming back to the λ-sequence as depicted in the summation terms in Eq. (13.1), following Littlewood
(1912) we shall suppose that the expression given in (13.2) is an ansatzk depicting the behavior of L(N ) for
large N .
The task that we then set ourselves, is to estimate the value of the exponent a in the asymptotic behavior
described in (13.2) |L(N )| = C N a which involves the λ−sequence. We do know that the λ-sequence does not all
behave like coin tosses, but we have shown that there exist subsequences of λ’s that exhibit a close correspondence
to the statistical distribution of coin tosses and though such subsequences are of relatively short lengths M (N ),
there are very many in number. Now a ‘True’ value of the exponent ‘a’ should be able to capture the correct
statistics in all such subsequences and predict the behavior of coin tosses for such subsequences. We now
investigate if such a True value for a exists and, if so, what its value should be. √
We will estimate the contribution to L(N ) for the same subsequence (5) of length M (N ) = N , then the
P when recomputed with an exponent a 6= 1/2 would give Pa :
N
X
Pa = λ(n) (13.50 )
√
n=N − N +1

That is √
Pa = L(N ) − L(N − N )
√
= CN a − C(N − N )a (11)
Simplifying by using Binomial expansion we have:
1 a(a − 1) a−1
Pa = CaN a− 2 − C N (12)
1.2

k Eq. C1 C2
(13.2) can be thought of as the first term in the asymptotic expansion of L(N ) for large N i.e. |L(N )| = N a (C + N
+ N2
+ ...)

Page 176
22 K. Eswaran

From the properties of the λ’s deduced from earlier results in this paper (Theorem 3, Appendices 3,4 and
Theorem A5, page 21), we now know that in actuality the particular subsequence in Eq.(13.5) and Eq.(13.50 )
contain random values of +1 and −1 and since the subsequence of λ’s have the same statistics as those of coin
tosses, Pa must be similar to P1/2 . Thus from (13.10) and (13.12)

Pa = O(1). (13)
From (13.11) this means that
1
CaN a− 2 = O(1) (14)
Since N is arbitrary and very large, this is impossible unless the condition
1
a= (15)
2
strictly holds.∗∗
††
Hence we have proved
√ a = 1/2. Since for consistency , condition (15), which arises from (13), is mandatory
and therefore |L(N )| ∼ N describes the asymptotic behavior of the summatory Liouville function. 

13 APPENDIX VI
On Coin Tosses and the Proof of Riemann Hypothesis
This Appendix has been written in such a manner that it can be read as a supplement to the
main paper and the first five appendices.
In the main part of this paper and the forgoing appendices, which we denote as: [MP and A’s], we
had proved the validity of the Riemann Hypothesis (RH). In this Appendix (VI), we perform a numerical
analysis and provide supporting empirical evidence that is consistent with the formal theorems that were key to
establishing the correctness of the RH. In particular, the numerical results of the statistical tests performed here
are firmly consistent with the proposition (formally proved in the paper cited above) that the values taken on
by the Liouville function over large sequences of consecutive integers are random. By performing this exhaustive
numerical analysis and statistical study we feel that we have provided a clearer understanding of the Riemann
Hypothesis and its proof.
1. Introduction
The Riemann zeta function, ζ(s), is defined by
∞
X 1
ζ(s) = , (1)
ns
n=1

where n is a positive integer and s is a complex variable, with the series being convergent for Re(s) > 1. This
function has zeros (referred to as the trivial zeros) at the negative even integers −2, −4, . . .. It has been shown‡‡
that there are an infinite number of non-trivial zeros on the critical line at Re(s) = 1/2. Riemann’s Hypothesis
(RH), which has long remained unproven, claims that all the nontrivial zeros of the zeta function lie on the
critical line. The main paper contains the proof [MP and A’s]
In this technical note, we provide a more concrete understanding and appreciation of the steps involved
in the proof of the Riemann Hypothesis by supplying supporting empirical evidence for those various theorems
which were proved and which had played a key role in the proof of the RH. In what follows we first give a brief
summary of how the RH was proved in [MP and A’s]. The proof followed the primary idea that if the zeta
function has zeros only the critical line, then the function F (s) ≡ ζ(2s)/ζ(s) cannot be analytically continued
to the left from the region Re(s) > 1, where it is analytic, to the left of Re(s) < 1/2. This point was recognized
by Littlewood as far back as 1912.∗ The function F (s) can be expressed as (see Titchmarsh (1951, Ch. 1)):

∗∗ In the above we tacitly assumed that a > 1/2, but a < 1/2 is not possible because then Pa will become zero. This implies that
dL/dN = 0, meaning |L(N )| will be a constant. But this again is impossible from Theorem 1, which would imply that F (s) can be
analytically continued to Re(s) = 0—an impossibility because of the presence of an infinity of zeros at Re(s) = 1/2, first discovered
by Hardy. √
†† It may be noted that for every (large) N there is a set S , Eq (13.4), containing M = N consecutive integers whose λ-values
N
behave like coin tosses; but there are an infinite number of integers N and therefore there are an infinite number of sets SN , for
which (13) must be satisfied.
‡‡ This was first proved by Hardy (1914).
∗ It may be noted that Littlewood studied the function 1/ζ(s) whereas we, in our analysis study F (s) ≡ ζ(2s)/ζ(s). This has made
things simpler.

Page 177
 A Proof of RH 23

∞
X λ(n)
F (s) = , (2)
ns
n=1

where λ(n) is the Liouville function defined by λ(n) = (−1)ω(n) , with ω(n) being the total number of prime
numbers in the factorization of n, including the multiplicity of the primes. The proof of RH in [MP and A’s]
requires also the summatory Liouville function, L(N ), which is defined as:
N
X
L(N ) = λ(n) (3)
n=1

The proof crucially depends on showing that the function F (s) = ζ(2s)/ζ(s), has poles only on the critical
line s = 1/2 + iσ, which translates to zeros of ζ(s), on the self same critical line s = 1/2 + iσ, because all the
values of s which appear as poles of F (s) are actually zeros of ζ(s), except for s = 1/2. Since, the trivial zeros
of ζ(s) which occur at s = −2, −4, −6.... that is negative even integers, conveniently cancel out from numerator
and denominator of the expression in F (s)), leaving only the non trivial zeros, also the pole of ζ(2s) will appear
as a pole of F (s), at s = 1/2. So it just remains to show that all the poles of F (s) lie on the critical line. This
was the Primary task of the paper.
The crucial condition then is that F (s) is not continuable to the left of Re(s) < 1/2, and therefore that
the zeta function have zeros only on the critical line,† is that the asymptotic limit of the summatory Liouville
function be |L(N )| ∼ C N 1/2 , where C is a constant. Therefore, to provide a rigorous proof of the validity
of the Riemann Hypothesis, [MP and A’s] investigated the asymptotic limit of L(N ). The work involved the
establishment of several relevant theorems, which were then invoked to eventually prove the RH to be correct.
We now state some of these important theorems‡ ).
Theorem 1:
In the set of all positive integers, for every integer which has an even number of primes in its factorization
there is another unique integer (its twin) which has an odd number of primes in its factorization.
Remark: Theorem 1 gives us the formal result that P r(λ(n) = +1) = P r(λ(n) = −1) = 1/2 , where Pr
denotes probability. That is, the λ-function behaves like an ‘ideal coin’.

Theorem 2:
Consider the sequence S− (N ) comprising µ(N ) consecutive positive integers, √ defined by
S− (N ) = {N − µ(N ) + 1, N − µ(N ) + 2, N − µ(N ) + 3, ..., N }, where µ(N ) = N . Then every number in
S− (N ) will differ in its prime factorization from that of every other number in S− (N ) by at least one distinct
prime.§
Remark: It is not hard to prove that the sequence S+ (N ) ≡ {N + 1, N + 2, N + 3, ........, N + µ(N )} of
length µ(N ) also behaves similarly. That is, every member of S+ (N ) differs from every other member by at least
one prime in its factorization. This implies, interestingly, that by choosing N to be consecutive perfect squares,
the entire set of positive integers can be envisaged as a union of mutually exclusive sequences like S− (N ) and
S+ (N ).
λ
It follows that the λ-values in the sequences S− (N ) ≡ {λ(N − µ(N ) + 1), λ(N − µ(N ) + 2), ..., λ(N )} and
λ
S+ (N ) ≡ {λ(N + 1), λ(N + 2), λ(N + 3), ..., λ(N + µ(N ))}, in which each member has a value either +1 or −1,
would also appear randomly and be statistically similar to sequences of coin tosses.
λ λ λ
Since the number of members in the sequences S− (N ), S− (N ), S+ (N ), and S+ (N ) is given by µ(N ) =
√
N → ∞ as N → ∞, the behavior of the λ-values of very large integers should coincide with that of a sequence
of coin tosses. This intuition was formally confirmed in Appendix V .
Theorem 3:
The summatoryqLiouville function takes the asymptotic form |L(N )| = C N 1/2 ,C is a constant. It can
2
be shown that C = π. It may be mentioned here that Littlewood’s condition is fairly tolerant: As long as
asymptotically, for large N , |L(N )| = C N 1/2 , and C is any finite constant, R.H. follows. This ‘tolerance’ is
reflected in the value of χ2 (below) as may be deduced, after a study of the following.
Remark: The form of the summatory Liouville function in Theorem 3 is precisely what we would expect for
a sequence of unbiased coin tosses. This, along with a sufficient condition derived by Littlewood (1912), shows

† Riemann had already shown that symmetry conditions ensure that there will be no zeros 0 < Re(s) < 1/2 if it is found that there
are no zeros in the region 1/2 < Re(s) < 1
‡ In addition to the theorems given below, a necessary theorem which states that: The sequence λ(1), λ(2), λ(3), ..., λ(n), ..., is
asymptotically non-cyclic, (i.e. it will never repeat), was also proved, in [MP and A’s], the theorems are numbered differently
§ For example, if two numbers c and d in S are factorized as c = pe1 pe2 and d = pe3 pe4 then at least one of the primes p or p will
1 2 3 4 3 4
be different from p1 or p2 .

Page 178
24 K. Eswaran

that F (s) is analytic for Re(s) > 1/2 and Re(s) < 1/2, thereby leaving the only possibility that the non-trivial
zeros of ζ(s) can occur only on the critical line Re(s) = 1/2.
In the following sections, by comparing the λ-sequences obtained for large sets of consecutive integers with
(binomial) sequences of coin tosses, we show that the statistical distributions of the two sets of sequences are
consistent with the claims of the above theorems. To this end, we apply Pearson’s ‘Goodness of Fit’ χ2 test.
The software program Mathematica developed by Wolfram has been used in this technical report to aid in the
prime factorization of the large numbers that this exercise entails.
The compelling bottom line that emerges from this empirical study is that it is extremely unlikely, in fact
statistically impossible, that for large N , the sequences of λ-values can differ from sequences of coin tosses. It is
this behavior of the Liouville function, recall, that delivers Theorem 3 above. And this Theorem, in turn, nails
down all the non-trivial zeros of the zeta function to the critical line [MP and A’s].

2. χ2 Fit of a λ-Sequence
In this section we will derive an expression of how closely a λ sequence corresponds to a binomial sequence
(coin tosses). We follow the exposition given in Knuth (1968, Vol. 2, Ch. 3); and then derive a very important
expression for a χ2 fit of a λ-Sequence, given by Eq.(14.9) below.
Suppose we are given a sequence, T (N0 , N ), of N consecutive integers starting from N0 :
T (N0 , N ) = {N0 , N0 + 1, N0 + 2, N0 + 3, ......., N0 + N − 1}
and the sequence, Λ(N0 , N ), of the corresponding λ-values:
Λ(N0 , N ) = {λ(N0 ), λ(N0 + 1), λ(N0 + 2), λ(N0 + 3), .. , λ(N0 + N − 1)}.
We ask how close in a statistical sense the sequence Λ(N0 , N ) is to a sequence of coin tosses or, in other
words, a binomial sequence. By identifying λ(n) = 1 as Head and λ(n) = −1 as T ail, for the nth ‘toss’, we
may perform this comparison. If this is really the case then statistically Λ(N0 , N ) should resemble a binomial
distribution, we can then compute the χ2 statistic as follows.

(P − EP )2 (M − EM )2
χ2 (N ) = + , (4)
EP EM
where P and M are the actual number of +1s (Heads) and −1s (T ails), respectively, in the Λ(N0 , N ) sequence,
EP and EM are the expectations of the number of +1s and −1s in the probabilistic sense. From Theorem 1 it
immediately follows that, for large N ,

EP = EM = N/2. (5)
We define L(N0 , N ) as the additional contribution to the summatory Liouville function of N consecutive
integers starting from N0 :
+N −1
N0X
L(N0 , N ) = λ(n). (6)
n=N0

b ≡ L(N0 , N ) and since (6) contains P terms which are equal to +1s and M
For brevity we will denote L
terms which are equal to −1s, we can write:

P − M = L,
b (7)
and

P + M = N. (8)
Using (7)and (8) we see that P = (N + L)/2b and M = (N − L)/2
b and from (5) we deduce P − EP = L/2
b
2
and M − EM = −L/2 b and thus equation (4) gives us the very important χ relation which is satisfied by every
Λ(N0 , N ) sequence involving the factorization of N consecutive integers starting from N0 :

[L(N0 , N )]2
χ2 (N ) = . (9)
N
Note that the it was possible to derive an expression for χ2 for largeN only because of Theorems
√ 1, 2, and
3. Now we particularly choose
√ N to be the square of an integer
√ and the sequence of length µ(N ) = N starting
from the integer N0 = N − N + 1 and then taking the N consecutive terms of the λ-sequence, we obtain
√ √ √ √
Λ(N0 , N ) = {λ(N − N + 1), λ(N − N + 2), λ(N − N + 3), ..., λ(N ) }, (10)

Page 179
 A Proof of RH 25

√ √
and the corresponding χ2 ( N ) for such a sequence (which is of length N ) can be obtained from Theorem 3
and (9) as
p√
√ [C N ]2
χ2 ( N ) = √
N
= C 2. (11)
√
Equation
√ (11) of course, should be interpreted as the average value of a sequence such as Λ(N0 , N ) of
length N given in the expression√(10). In this report we perform the χ2 ‘Goodness of Fit’ tests for very
many sequences of the type Λ(N0 , N ) with varying lengths and very large values of N to examine whether
these sequences are statistically indistinguishable from coin tosses. In this manner, we provide empirical support
for the claims of the theorems formally proved in [MP and A’s] and, therefore, for the proof of the Riemann
Hypothesis.
√
3. Numerical Analysis of Sequence √Λ(N0 , N ) and its χ√2 Fit
In this section, we consider sequences of length N , starting from N0 = N − N + 1 or N + 1 where N
is a perfect square. We use Mathematica to compute L(N0 , N ).¶
In the table below we list the sequences in the following format. We define the sequences:
√ √
S− (N ) = {N − N + 1, N − N + 2, ..., N }, (12)

√
S+ (N ) = {N + 1, N + 2, ..., N + N }, (13)
and the partial sums of the λs of the two sequences defined above are defined by the expressions:
√ √ √
L(S− ) ≡ L(N − N + 1, N ) = λ(N − N + 1) + λ(N − N + 2) + ... + λ(N ), (14)

√ √
L(S+ ) ≡ L(N + 1, N + N ) = λ(N + 1) + λ(N + 2) + ... + λ(N + N ). (15)
The√ formal √proof of the Riemann √ Hypothesis in [MP and A’s] proceeded as follows. The sequences
Λ(N − N √ + 1, N ) and Λ(N + 1, N ) were shown to behave like coin tosses for every N (large) over sequences
of length N , where N is taken to be a perfect square. On taking N to be consecutive perfect squares, the
lengths of the consecutive sequences naturally increase. Using this procedure, we obtain sequences that can span
the entire set of positive integers (consult the first five columns of Tables 1.1 to 1.4). Since the λs within each
segment behave like coin tosses, from√ the work of Chandrashekar (1943) it follows that the summatory Liouville
function L(N ) must behave like C N as N → ∞. The validity of RH follows, by Littlewood’s Theorem, from
the fact that F (s) cannot then be continued to the left of the critical line Re(s) = 1/2 because of the appearance
of poles in F (s) on the line, each pole corresponding to a zero of the zeta function ζ(s).

Statistical Tests
We shall now test the following null hypothesis H0 against the alternative hypothesis H1 in the following
generic forms:
H0 : The sequence Λ(N0 , N ) has the same statistical distribution as a corresponding sequence of coin tosses
(i.e. binomial distribution with P rob(H) = P rob(T ) = 1/2).
H1 : The sequence Λ(N0 , N ) has a different statistical distribution than a corresponding sequence of coin
tosses (i.e. binomial distribution with P rob(H) = P rob(T ) = 1/2).
The critical value for chi square is χ2crit = 3.84, for the standard 0.05 level of significance. (In our case,
the relevant degrees of freedom equal to 1.) Assuming that H0 is true, if chi square is less than χ2crit the null
hypothesis is accepted.

It should be noted that the tests conducted here are not merely exploratory statistical exercises to discern
possible patterns in the λ-sequences. Rather, the tests here are informed by theory. We have formally shown in
[MP and A’s] that, over the set of positive integers, the probability that λ takes on the value +1 or −1 with

¶A
PK
typical Mathematica command which calculates the expression n=J λ(n) is:
Plus[LiouvilleLambda[Range[J,K]]].For instance, the command which sums
the λ(n) from n = 25, 000, 001 to 25, 005, 000 is:
Plus[LiouvilleLambda[Range[25000001, 25005000]]], which will give the answer = −42.

Page 180
26 K. Eswaran

equal probability and that, over sequences that are increasing in N , the λ draws are random. Thus statistical
evidence consistent with these claims√ merely bolster what has already been formally demonstrated.
The behavior of the Λ(N0 , N ) sequences are verified to be indeed like coin tosses for a very large number
of cases and the results are summarized in the tables below. Let us take an example from Table 1.1. The third
row gives the χ2 result for the sequence of length 1001, starting from 1001001. We can factorize each of these
numbers as:
1001001 = 3 × 333667; 1001002 = 2 × 500501; 1001003 = prime;
1001004 = 22 × 3 × 83417;.........., 1001999 = 41 × 24439;
1002000 = 24 × 3 × 53 × . 167; 1002001 = 72 × 112 × 132
and hence we can evaluate the corresponding λ-sequence, by using the definition λ(n) = (−1)ω(n) , with
ω(n) being the total number of prime numbers (multilpicities included) in the factorization of n. We find that:
Λ(1001001, 1001) = {λ(1001001), λ(1001002), ..., λ(1002000), λ(1002001)}
= {1, 1, −1, 1, 1, 1, 1, −1, 1, −1, 1, ................, 1, −1, 1}.
The partial sum of all the 1001 λs shown in the sequence above adds up to 49. We then estimate how close
the sequence Λ(1001001, 1001) is to a Binomial distribution, i.e. of 1001 consecutive coin tosses. The observed
value χ2 = 2.4 for this sequence of λs is well below the critical value χ2crit = 3.84 (for a one degree of freedom)
at the standard significance level of 0.05. Thus the sequence Λ(1001001, 1001) is statistically indistinguishable
from a Binomial distribution obtained by 1001 consecutive coin tosses if we consider Head = +1 and T ail = −1.
In fact, it so happens that out of the 10 sequences shown in Table 1.1 this chosen example has the largest value
of χ2 ; the other sequences have a much lower χ2 value and the average value is 0.653 which hovers around the
predicted average C 2 = π2 = 0.637. We see that the null hypothesis would be accepted even if the significance
level were at 0.10, for which χ2crit = 2.71.
We have calculated the χ2 for larger and larger sequences see Tables1.2, Tables 1.3 and Tables 1.4 for √ even
very large numbers ∼ 1010 and sequences involving 105 consecutive integers in each case the sequences Λ(N0 , N )
behave like coin tosses thus lending emphatic empirical support consistent with the Theorems proved in [MP
and A’s],.
TABLE 1.1 Sequence of√Consecutive Integers of Type S− (N ) and S+ (N ) of Length 1000
No T ype of S(N) N From to L(S) χ2
1. S 1000 999,001 1,000,000 6 0.036
2. S+ 1000 1,000,001 1,001,000 10 0.100
3. S 1001 1,001,001 1,002,001 49 2.400
4. S+ 1001 1,002,002 1,003,002 -37 1.368
5. S 1002 1,003,003 1,004,004 -12 0.144
6. S+ 1002 1,004,005 1,005,006 -28 0.780
7. S 1003 1,005,007 1,006,009 3 0.009
8. S+ 1003 1,006,010 1,007,012 -39 1.516
9. S 1004 1,007,013 1,008,016 12 0.143
10. S+ 1004 1,008,017 1,009,020 6 0.036

MEAN χ2 FROM 999,001 1,009,020 = 0.653

TABLE 1.2 Sequence of√Consecutive Integers of Type S− (N ) and S+ (N ) of Length 5000
No T ype of S(N) N From to L(S) χ2
1. S 5000 24,995,001 25,000,000 0 0.0
2. S+ 5000 25,000,001 25,005,000 -42 0.353
3. S 5001 25,005,001 25,010,001 -27 0.148
4. S+ 5001 25,010,002 25,015,002 -103 2.12
5. S 5002 25,015,003 25,020,004 -76 1.155
6. S+ 5002 25,020,005 25,025,006 48 0.461
7. S 5003 25,025,007 25,030,009 -13 0.034
8. S+ 5003 25,030,010 25,035,012 119 2.831
9. S 5004 25,035,013 25,040,016 124 3.072
10. S+ 5004 25,040,017 25,045,020 62 0.768

MEAN χ2 FROM 24,995,001 25,045,020 = 1.094

Page 181
 A Proof of RH 27

TABLE 1.3 Sequence of Consecutive Integers of Type S− (N ) and S+ (N ) of Length 10,000

√
No T ype N From to L(S) χ2
1. S 10000 99,990,001 100,000,000 -146 2.132
2. S+ 10000 100,000,001 100,010,000 -88 0.774
3. S 10001 100,010,001 100,020,001 -11 0.012
4. S+ 10001 100,020,002 100,030,002 -43 0.185
5. S 10002 100,030,003 100,040,004 8 0.064
6. S+ 10002 100,040,005 100,050,006 36 0.130
7. S 10003 100,050,007 100,060,009 23 0.053
8. S+ 10003 100,060,010 100,070,012 -49 0.240
9. S 10004 100,070,013 100,080,016 -20 0.040
10. S+ 10004 100,080,017 100,090,020 112 1.254

MEAN χ2 FROM 99,990,001 TO 100,090,020 = 0.488

TABLE 1.4 Sequence of Consecutive Integers of Type S− (N ) and S+ (N ) of Length 100,000
√
No T ype N From to L(S) χ2
1. S 100,000 9,999,900,001 10,000,000,000 -232 0.538
2. S+ 100,000 10,000,000,001 10,000,100,000 340 1.15
3. S 100,001 10,000,100,001 10,000,200,001 -249 0.620
4. S+ 100,001 10,000,400,005 10,000,500,006 -115 0.132
5. S 100,002 10,000,300,003 10,000,400,004 216 0.467
6. S+ 100,002 10,000,400,005 10,000,500,006 456 2.08
7. S 100,003 10,000,500,007 10,000,600,009 -255 0.650
8. S+ 100,003 10,000,600,010 10,000,700,012 -235 0.552
9. S 100,004 10,000,700,013 10,000,800,016 -44 0.0194
10. S+ 100,004 10,000,800,017 10,000,900,020 202 0.408
11. S 100,005 10,000,900,021 10,001,000,025 -191 0.364
12. S+ 100,005 10,001,000,026 10,001,100,030 475 2.26
13. S 100,006 10,001,100,031 10,001,200,036 134 0.179
14. S+ 100,006 10,001,200,037 10,001,300,042 -66 0.0436
15. S 100,007 10,001,300,043 10,001,400,049 427 1.82
16. S+ 100,007 10,001,400,050 10,001,500,056 -303 0.918
17. S 100,008 10,001,500,057 10,001,600,064 276 0.762
18. S+ 100,008 10,001,600,065 10,001,700,072 -210 0.441
19. S 100,009 10,001,700,073 10,001,800,081 267 0.713
20. S+ 100,009 10,001,800,082 10,001,900,090 291 0.847

MEAN χ2 FROM 9,999,900,001 TO 10,001,900,090 = 0.768

3.1 Sequences of Fixed Length Arbitrarily Positioned

In this section we consider various segments of consecutive integers of a fixed length but starting from an
arbitrary integer. Even here we see that the λs within each segment behave like coin tosses and have the same
statistical properties.
We now calculate the χ2 values of λ-sequences for a sequence SA of consecutive integers, starting from an
arbitrary number N0 but all of a fixed length M :

SA (N ) = {N0 , N0 + 1, N + 2, N + 3, ..., N0 + M − 1} (16)

and
L(SA ) = λ(N0 ) + λ(N0 + 1) + λ(N0 + 2) + λ(N0 + 3) + ... + λ(N0 + M − 1) (17)
The results, which are summarized in Table 2.1, again show that the λ-sequences are statistically like coin tosses.

Page 182
28 K. Eswaran

TABLE 2.1 Sequence of Consecutive Integers of Type SA (N ) and of Length M = 1000

No T ype M From N0 to N0 + M − 1 L(SA ) χ2
1. SA 1000 10,000,001 10,001,000 36 1.296
2. SA 1000 12,000,001 12,001,000 28 0.784
3. SA 1000 13.000,001 13,001,000 -14 0.196
4. SA 1000 15,000,001 15,001,000 10 0.10
5. SA 1000 45,000,001 45,001,000 -18 0.324
6. SA 1000 47,000,001 47,001,000 -36 1.296
7. SA 1000 56,000,001 56,001,000 24 0.576
8. SA 1000 70,000,001 70,001,000 -44 1.936
9. SA 1000 90,000,001 90,001,000 14 0.196
10. SA 1000 95,600,001 95,601,000 28 0.784
11. SA 1000 147,000,001 147,001,000 -26 0.676
12. SA 1000 237,000,001 237,001,000 -24 0.576
13. SA 1000 400,000,001 400,001,000 26 0.676
14. SA 1000 413,000,001 413,001,000 10 0.10
15. SA 1000 517,000,001 517,001,000 14 0.196
16. SA 1000 530,000,001 530,001,000 -32 1.024
17. SA 1000 731.000,001 731,001,000 50 2.500
18. SA 1000 871,000,001 871,001,000 -42 1.764
19. SA 1000 979,000,001 979,001,000 -20 0.400
20. SA 1000 997,000,001 997,001,000 14 0.196

MEAN χ2 OF ABOVE 20 SEGMENTS = 0.780

3.2 Entire Sequences from n = 1 to n = N , N large and calculation of χ2 for

such sequences from L(N )
PN
It has been empirically verified in the literature that the summatory Liouville Function L(N ) = n=1 λ(n)
fluctuates from positive to negative values as N increases without bound. We now investigate the χ2 values for
such sequences,and use Eq.(9), so that we may see how these sequences behave like coin tosses.
In the Table 3.1 we use the values of L(N ) for various large values of N , which were found by Tanaka
(1980), the results depicted below reveal that the lambda sequences are statistically indistinguishable from the
sequences of coin tosses over such large ranges of N from 1 to one billion.
In the above we calculated L(N ) for various valies of N , however, if we choose a value N at which L(N ) is
a local maximum or a local minimum then we would be examining potential worst case scenarios for deviations
of the λs from coin tosses because these are the values of N that are likely to yield the highest values of χ2 (see
equation (9)). It is interesting to investigate if even for these special values of N whether the χ2 is less than the
critical value; if so, we would again have statistical assurance that the entire sequence of λs from n = 1, 2, 3, ...
behave like coin tosses.
We therefore use the 58 largest values of L(N ) and the associated values of N reported in the literature
by Borwein, Ferguson and Mossinghoff (2008), and perform our statistical exercise. See Table 3.2. We see that
even for these “worst case scenario” values of N the lambda sequences are statistically indistinguishable from
the sequences of coin tosses.

Page 183
 A Proof of RH 29

TABLE 3.1 Values of L(N) at various large values of N

(The values for N and L(N) are from Tanaka (1980))
PN
No. N L(N ) = n=1 λ(n) χ2
1 100,000,000 -3884 0.1508
2 200,000,000 -11126 0.6189
3 300,000,000 -16648 0.9238
4 400,000,000 -11200 0.3136
5 500,000,000 -18804 0.7072
6 600,000,000 -15350 0.3927
7 700,000,000 -25384 0.9204
8 800,000,000 -19292 0.4652
9 900,000,000 -4630 0.0238
10 1,000,000,000 -25216 0.6358

MEAN χ2 OF ABOVE = 0.5152

TABLE 3.2 Values of L(N) at local Minima (Maxima) for very Large N
(The values for N and L(N) are from Borwein, Ferguson and Mossinghoff (2008))
PN
No. N L(N ) = n=1 λ(n) χ2
1 293 -21 1.5051
2 468 -24 1.2308
3 684 -28 1.1462
4 1,132 -42 1.5583
5 1,760 -48 1.3091
6 2,804 -66 1.5535
7 4,528 -74 1.2094
8 7,027 -103 1.5097
9 9,840 -128 1.665
10 24,426 -186 1.4164
11 59,577 -307 1.582
12 96,862 -414 1.7695
13 386,434 -698 1.2608
14 614,155 -991 1.5991
15 925,985 -1,253 1.6955
16 2,110,931 -1,803 1.54
17 3,456,120 -2,254 1.47
18 5,306,119 -2,931 1.619
19 5,384,780 -2,932 1.5965
20 8,803,471 -3,461 1.3607

Page 184
30 K. Eswaran

TABLE 3.2 (Cont’d) Values of L(N) at local Minima (Maxima)

PN
No. N L(N ) = n=1 λ(n) χ2
21 12,897,104 -4,878 1.845
22 76,015,169 -10,443 1.4347
23 184,699,341 -17,847 1.7245
24 281,876,941 -19,647 1.3694
25 456,877,629 -28,531 1.7817
26 712,638,284 -29,736 1.2408
27 1,122,289,008 -43,080 1.6537
28 1,806,141,032 -50,356 1.4039
29 2,719,280,841 -62,567 1.4396
30 3,847,002,655 -68,681 1.2262
31 4,430,947,670 -73436 1.2171
32 6,321,603,934 -96,460 1.4719
33 10,097,286,319 -123,643 1.514
34 15,511,912,966 -158,636 1.6223
35 24,395,556,935 -172,987 1.2266
36 39,769,975,545 -238,673 1.4324
37 98,220,859,787 -365,305 1.3586
38 149,093,624,694 -461,684 1.4296
39 217,295,584,371 -598,109 1.6463
40 341,058,604,701 -726,209 1.5463
41 576,863,787,872 -900,668 1.4062
42 835,018,639,060 -1,038,386 1.2913
43 1,342,121,202,207 -1,369,777 1.398
44 2,057,920,042,277 -1,767,635 1.5183
45 2,147,203,463,859 -1,784,793 1.4836
46 3,271,541,048,420 -2,206,930 1.4888
47 4,686,763,744,950 -2,259,182 1.089
48 5,191,024,637,118 -2,775,466 1.4839
49 7,934,523,825,335 -3,003,875 1.1372
50 8,196,557,476,890 -3,458,310 1.4591
51 12,078,577,080,679 -4,122,117 1.4068
52 18,790,887,277,234 -4,752,656 1.2021
53 20,999,693,845,505 -5,400,411 1.3888
54 29,254,665,607,331 -6,870,529 1.6136
55 48,136,689,451,475 -7,816,269 1.2692
56 72,204,113,780,255 -11,805,117 1.9301
57 117,374,745,179,544 -14,496,306 1.7904
58 176,064,978,093,269 -17,555,181 1.7504

The empirical evidence provided here is very comprehensive: it examines the statistical behavior of the
Liouville function for large segments of consecutive integers (e.g.Table 1.4). We have also considered the
entire series of λ(n) from the values of n = 1 to n = N = 176 trillion - as high as any available studies in
the literature have gone. And yet, the λ−sequences consistently show themselves, in rigorous statistical tests, to
be indistinguishable from sequences of coin tosses, hencepproviding overwhelming statistical evidence in support
of Littlewood’s condition that as N → ∞, L(N ) = C. (N ), (where C is finite) and thus declaring that the
non-trivial zeros of the zeta function, ζ(s), must all necessarily lie on the critical line Re(s) = 1/2.
4. Concluding Note
In this Appendix VI we have provided compelling, comprehensive numerical and statistical evidence that
is consistent with the Theorems that were instrumental in the formal validation of the Riemann Hypothesis in
[MP and A’s].
It is hoped that a perusal of this section (Appendix 6)report offers some insight into, and understanding of,
why the Riemann Hypothesis is correct. It should be noted that, while the results presented here are perfectly
consistent with the theoretical results in [MP and A’s], they obviously do not prove (in a strict mathematical
sense, because of the statistical nature of the study), the Riemann Hypothesis. For the formal proof, the rigorous
mathematical analysis in the main paper needs to be consulted.

Page 185
 A Proof of RH 31

14 References
1. Apostol, T.,(1998) Introduction to Analytic Number Theory, Chapter 2, pp. 37-38, Springer International,
Narosa Publishers, New Delhi.

2. Balasubramanian, R., Laishram, S., Shorey, T.N. and Thangadurai,R.,(2009) ‘The Number of Prime
Divisors of a Product of Consecutive Integers’, J. of Combinatorics and Number Theory, 1, no 3, pp 253-261

3. Borwein, P., Choi,S., Rooney,B., and Weirathmueller,A., (2006) The Riemann Hypothesis, Springer. Also
see slides 28 to 32 in Lecture:
http://www.cecm.sfu.ca/personal/pborwein/SLIDES/RH.pdf

4. Chandrasekhar S., (1943)’Stochastic Problems in Physics and Astronomy’,Rev.of Modern Phys. vol 15,
no 1, pp1-87

5. Denjoy, A., (1931) L’ Hypothese de Riemann sur la distribution des zeros.. C.R. Acad. Sci. Paris 192,
656-658

6. Edwards, H.M., (1974), Riemann’s Zeta Function, Dover Publications, New York.

7. Eswaran, K., (1990)‘On the Solution of Dual Integral Equations Occurring in Diffraction Problems’,
Proc. of Roy. Soc. London, A 429, pp. 399-427

8. Eswaran, K, (Sept 2016 ) and (Rev. 2017) ‘The Dirichlet Series for the Liouville Function and the
Riemann Hypothesis’, https://arxiv.org/pdf/1609.06971.pdf,
Comment by Author: These papers in Arxiv and the other preprints in Researchgate, which
bear the same title, are older versions of the present paper and are probably harder to understand

9. Eswaran, K., (April, 2018) ‘A Simple Proof That Even and Odd Numbers of Prime Factors Occur
with Equal Probabilities in the Factor-ization of Integers and its Implications for the Riemann Hypothesis.
Rsearchgate Preprint www.researchgate.net/publication/324828748

10. Grimm, C.A., (1969) ‘A conjecture on consecutive composite numbers’,Amer. Math Monthly, 76,
pp1126-1128.

11. Hardy, G.H., (1914), ‘Sur les zeros de la fonction ζ(s) de Riemann,’ Comptes Rendus de l’Acad. des
Sciences (Paris), 158, 1012–1014.

12. Khinchine, A. (1924) “Uber einen Satz der Wahrscheinlichkeitsrechnung”, Fundamenta Mathematicae
6, pp. 9-20

13. Knuth D.,(1968) ‘Art of Computer Programming’, vol 2,Chap 3. Addison Wesley

14. Kolmogorov, A., (1929) “Uber das Gesetz des iterierten Logarithmus”. Mathematische Annalen, 101:
126-135. (At the DigitalisierungsZentrum web site)). Also see: https:en.wikipedia.orgwikiLaw of the iterated
logarithm.

15. Laishram, S. and Ram Murty(2012), ‘Grimm’s conjecture and smooth numbers’, Michigan Math.J. 61,
pp 151-160

16. Laishram, S. and Shorey, T.N., (2006), ‘Grimm’s Conjecture on Consecutive integers’, Int. J. of Number
Theory, 2, 207-211.

17. Landau,E., 1899 Ph. D Thesis, Friedrich Wilhelm Univ. Berlin

18. Littlewood, J.E. (1912), ‘Quelques consequences de l’hypoth‘ese que la fonction ζ(s) de Riemann n’a
pas de zeros dans le demi-plan R(s) > 1/2,’ Comptes Rendus de l’Acad. des Sciences (Paris), 154, pp. 263–266.
Transl. version,in Edwards op cit.

Page 186
 32 K. Eswaran

19. Press,W.H, Teukolsky,S.A., Vetterling, W.T. and Flannery,B.P., Press etal (1986)‘On Numerical
Recipes in Fortran’,Chap 7, Cambridge univ. Press, 1986

20. Ramachandra, K., Shorey T.N.and Tijdeman, R. (1975), ‘On Grimm’s problem relating to factorization
of a block of consecutive integers’,I and II Reine Angew. Math. 273, pp 109-124; 288 192-201.

21. Riemann,B., (1859)in ‘Gessammeltz Werke’, Liepzig, 1892, trans. ‘On the number of primes less than
a given magnitude’ available In Edwards’ book, opcit pp299-305

22. Tanaka, Minoru, (1980), ‘A numerical investigation of the cumulative sum of the Liouville function’,
Tokyo J. of Math. vol 3, No. 1, pp. 187-189.

23. Titchmarsh, E.C. (1951), The Theory of the Riemann Zeta-Function, Clarendon Press, Oxford.

24. Whittaker, E.T. and Watson, G.N., (1989)‘A Course of Modern Analysis’, Universal Book Stall New
Delhi

Page 187
View publication stats
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/331889126

The Pathway to the Riemann Hypothesis

Preprint · March 2019

DOI: 10.13140/RG.2.2.17950.59208

CITATIONS READS
0 190

1 author:

K. Eswaran
Sreenidhi Institute of Science & Technology
87 PUBLICATIONS   325 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The Dirichlet Series for the Liouville Function and the Riemann Hypothesis View project

Development of New Pattern Recognition Algorithms for use in classification and Neural network Research View project

All content following this page was uploaded by K. Eswaran on 20 March 2019.

The user has requested enhancement of the downloaded file. Page 188
 The Pathway to the Riemann Hypothesis
K. Eswaran

March 20, 2019

This brief Note describes the scheme and the essential steps in the proof of
the RH as given by K.Eswaran in his Main Paper.:
The Final and Exhaustive Proof of The Riemann Hypothesis from First
Principles. See Ref [1] (in Reference List below).
This write up also provides an answer to queries raised by readers and other
researchers.

The proof involves ve basic steps which are descibed below:

1 Step 1:
To look for an analytic function, F(s), whose poles exactly correspond to the non-
trivial zeros of the zeta function ζ(s), so that various mathematical techniques
discovered in the eld of complex function theory can be used to analize F(s).
It was then found that the function F(s) which is dened as below:

ζ(2s)
F (s) = (1)
ζ(s)
has all the above required properties: F(s) has poles at exactly the same
positions as ζ(s) has its non-trivial zeros in the critical region, (the trivial zeros,
s = −2n (n: an integer), cancel out and do not appear as poles in F(s); also for
F(s), there is one additional pole which corresponds to the single pole in ζ(2s),
but this does not disturb or play any role in our analysis, because it occurs any
way on the critical line s = 1/2).
F(s) is analytic in the region Re(s) > 1and is given by:
∞
X λ(n)
F (s) = (2)
n=1
ns

2 Step 2:
In this step the necessary and sucient conditions for the analyticity of F(s)
is determined. When a technique used by Littlewood for determining the be-
haviour of the function 1/ζ(s) by analytic continuation is used to nd the be-
haviour of F(s). It will then be demonstrated that the summatory function L(N)

Page 189
which is dened by:
N
X
L(N ) = λ(n) (3)
n=1

plays a crucial role in determining the position of the poles of F(s) and
thereby the zeros of ζ(s) in the critical region 0< Re(s) <1. Littlewood's theo-
rem states that the asymptotic behaviour of L(N ) for large N , determines the
analyticity of F(s), and if the behaviour is such that

N
X
| L(N ) | ≡ | λ(n) |< C N a+ (f or large N ) (4)
n=1

(where (1/2 ≤ a < 1 ), and  is a small positive number), F(s) will be analytic
in the region (a < Re(s). This is a very crucial result as far as RH is concerned
because if one can determine that actually a = 1/2 in (4) then the Riemann
Hypothesis is proved.

3 Step 3:
In this step we logically put forth the argument: that the very necessity that
(4) must be satised for the Riemann Hypothesis to be true, imposes very se-
vere resrtictions on the behaviour of the sequence of the Liouville functions:
{λ(1), λ(2), λ(3), .....}. These restrictions (conditions) will be delineated later in
this section.
The λ(n) is dened as:λ(1) = 1and for n > 1 : λ(n) = (−1)Ω(n) and
is determined by factorizing n and nding Ω(n), the number of prime factors
of n (multiplicities included). We already know λ(n) is fully determined by
factorizing n and is an arithmetic function namely: λ(m.n) = λ(m).λ(n), for all
integers m, n.
Now for RH to be true one must have a = 1/2 in Eq.(4), the rst N terms
(N large) of the λ sequence must therefore sum up as:

| λ(1) + λ(2) + λ(3) + ...... + λ(N ) | ' C . N 1/2 (5)

The above equation brings to mind a similar relationship satised by another

sequence c(n) = ±1, which corresponds to the nth step of a One-dimensional
random walk! (This c(n) can be simulated by coin tosses, if we replace Heads
by +1 and Tails by -1; so a N-step random walk can be thought as a coin toss
experiment where a coin is tossed N times.) It is well known that for such
a random-walker's sequence the sum indicates the distance travelled from the
starting position in N steps and satises the relationship:

| c(1) + c(2) + c(3) + ...... + c(N ) | ' C . N 1/2 (6)

Equation (6) is derived by using the assumption that the random walker
behaves in such a manner that:

Page 190
 (i) Each step is of the same size but can be either in the positive direction
or negative direction i.e the nth step c(n) can be +1 or -1,
with Equal Probability.
(ii) The sequence of steps cannot be periodic, that is the pattern of steps
cannot form a repetitive pattern.
(iii) Knowing the nth step the (n+1)th cannot be predicted. That is, knowing
c(n), c(n+1) cannot be determined (they are independent).

The above assumptions are enough to derive Eq(6). This has been shown
by many researchers (e.g. See S. Chandrasekar, refered in Ref[1])

3.1 The Argument:

Comparing (5) and (6) leads us to deduce some inevitable conclusions:

Eq.(5) must be satised by the λ(n) sequence if the Riemann Hypothesis
is TRUE, this is the conclusion that we deduce from Littlewoods Theorem.
However, (5) needs be satised only for large N (this being the condition of Lit-
tlewood's theorem). Now, there are many Random walks possible, for instance:
100 random walkers can each of them, take N steps and each of these random
walkers will be at anapproximate distance of distance C . N 1/2 from the starting
point. Each of these 100 sequences can be thought of as 100 dierent instances
of a random walk of N steps each.
If we wish to compare (6) with (5) there are several conceptual issues: (i) The
sequence in (5) is a deterministic sequence and (ii) we have only one sequence.
We get over this latter issue by considering the single sequence as one instance
of a hypothetical random walk of N steps. And even though the λ(n)'s are
deterministic (an aspect we temporarily ignore) we could investigate this one
instance and argue (or hypothesize) that when N is large, the following rules
could be obeyed:
Properties of the λ−sequence
(a) Given an arbitrary large n chosen at random, there is Equal probability
of λ(n)being either +1 or-1.
(b) The λ−sequence cannot be periodic, that is the λ(n) cannot form a
repetitive pattern (no cycle)
(c) Knowing the value of λ(n) it is not possible to predict λ(n + 1) . Unpre-
dictability (independence).

Note the rules (a),(b)and (c) are similar to (i),(ii) and (iii) and therefore: If
by using the number theoretical (arithmetical) properties of the inte-
gers, the primes and the factorization process, it is somehow possible
to prove that the λ(n) satisy the rules (a), (b) and (c) then just as (6)
is satised by every instance of a random walk, Eq(5) will be satised
for our one particular instance of our λ−sequence and thus RH will
be proved! Taking this as a cue we proceed.
Hence the next step is to prove the properties for large values of N i.e. when
N tends to innity. (It will become clear later that the deterministic nature of

Page 191
the λ(n)0 s, does not signicantly disturb the above statistical properties. .)
1

4 Step 4: Proof of the Properties of the λ−sequence

In this step several theorems are proved using the number theoretical (arith-
metical) properties of integers, primes and the unique factorization of integers
to establish the properties (a), (b) and (c) of the λ−sequence as listed in the
previous paragraphs. These proofs are fairly straight forward and are from rst
principles:
Property (a) On Equal Probabilities, is proved in Theorems 2 and 3 in
Section 5.2, in the Main Paper Ref [1]. The concept of towers is used in the
proofs. An alternative proof
2 by constuction of all prime products and induction
is also given in Ref [2]. A third proof, which follows from Littlewood's theorem
but assumes the fact that there is no zero with Re(s)=1, (proved in the Prime
Number Theorem) can also be derived (but is not given in the paper).
Property (b) On no cycles, is proved in Appendix III, Ref [1]
Property(c) On unpredictability (independence) is proved in Appendix IV,
Ref[1]. An alternative proof of this also given: See para 5(a), in page 2, of Ref
[3].
An alternative arithmetical proof of the asymptotic relation | L(N ) |' c =
C.N 1/2 . is given in Appendix V, Ref [1].
3

The above extablishes RH.

Finally, we show that the 'width' of the Critical Line' must necessarily tend
to zero. See: 3rd paragraph before the Conclusion in Ref[1].

5 Experimental verication
In the last Appendix VI, Ref[1], numerical experiments (using Mathematica)
are described and there it is shown that large sequence of lambdas behave like

1 For example, given λ(n) for some n, the formula λ(m.n) = λ(m).λ(n),can determine the
next predictable value λ(2.n) = λ(2).λ(n) = −λ(n), but for large n, say n = 10100 , the integer
2n will be at a distance of 10100 from n making such a prediction statistically insignicant!
2 This alternative proof of equal probabilities is given in Ref [2] and is done by explict
construction of integers by products of sets which are powers of a given prime. This is from
rst principles using the principle of mathematical induction and all that is assumed: is that
every odd number is succeded by an unique even number which is preceded by an unique odd
number.
3 In a separate arithmetical study Ref [4], it was discovered that for very large N, smaller
primes contribute more (than the larger primes) to the calculation of λ(n)'s which occur in the
summatory expression for L(N ). Specically, if one chooses an integer K such that K << N
then the primes p which are s.t p < N/K , occur much more often in the calculation of each
term in L(N ) than the large primes q which are s.t. N/K < q < N . This situation permits
us to deduce, interestingly, that if we allow both K and N → ∞ in such a manner that the
ratio N/K is a xed number, then we must have: P r(λ(n) = 1 | n < N ) = 1/2 − logN CK
and
P r(λ(n) = −1 | n < N ) = 1/2 + logN where CK is a small uctuating number which tends
CK

to zero as K → ∞; thus once again conrming that the L(N ) behaves like a random walk for
very large N.

Page 192
 a random walk (or equivalently like coin tosses). These sequences called S− (N )
and S+ (N ) exist (N being a perfect square) and behave like random sequences
(coin tosses) and the concatination of such sets of S− (N ) and S+ (N ) cover all
of λ(n) for all integers n upto innity. The Tables given in this Appendix VI,
provide ample proof of this, for instance, see Table 1.4. The purpose of this
section is just to demonstrate that the, predictions of the theorems proved have
been numerically veried extensively. The verication has been done by doing a
χ2 t of a λ−sequence with a Binomial distribution (coin tosses). In every case
it has been shown that for large N the λ−sequence is indistinguishable from a
random walk (sequence of coin tosses).

6 References
[1] The nal and Exhaustive proof of the Remann Hypothesis...

[2]A Simple Proof That Even and Odd Numbers of Prime Factors Occur
with Equal Probabilities in the Factor-ization of Integers

[3]A Quick Reading Guide to the Proof of the Riemann Hypothesis

[4] The eect of of the non-random-walk behavior of the Liouville Series

L(N) by the rst nite number of terms.

I enclose below the slides of the Invited Lecture that I delivered at the
Government Arts & Science College Kumbakonam on March 1st 2019. (This
was followed by another (slightly shorter) Lecture delivered in the Ramnujan
Centre of Sastra University on the evening of the same day).
[5] Invited Lecture On the Riemann Hypothesis by K.Eswaran

K. Eswaran/Professor
Sree Nidhi Institute of Science and Technology,
Yamnampet, Ghatkesar, Hyderabad 501301
20/3/2019

Page 193
View publication stats
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/324828748

A Simple Proof That Even and Odd Numbers of Prime Factors Occur with Equal
Probabilities in the Factor-ization of Integers and its Implications for the
Riemann Hypothesis

Preprint · April 2018

DOI: 10.13140/RG.2.2.28975.43684/2

CITATIONS READS

0 791

1 author:

K. Eswaran
Sreenidhi Institute of Science & Technology
87 PUBLICATIONS   325 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The Dirichlet Series for the Liouville Function and the Riemann Hypothesis View project

Development of New Pattern Recognition Algorithms for use in classification and Neural network Research View project

All content following this page was uploaded by K. Eswaran on 30 April 2018.

The user has requested enhancement of the downloaded file. Page 194
A Simple Proof That Even and Odd Numbers of Prime
Factors Occur with Equal Probabilities in the Factor-
ization of Integers and its Implications for the Riemann
Hypothesis
by K. Eswaran
Sreenidhi Inst. of Science and Technology, Yamnampet, Ghatkesar,Hyderabad
501301, INDIA

ABSTRACT: In this paper it is demonstrated that that the probability

that an integer has an even number of prime factors (multiplicity included)
is equal to the probability that it has an odd number of prime factors. This
result, which was already proved alternatively in [6], is closely connected to the
Riemann Hypothesis (RH) and provides a route to the proof of RH. The method
followed in [6] to proving RH is briefly outlined.

1 The Result
In the following, we construct the integers from the prime numbers by reverse
engineering the Prime Factorization Theorem. This provides a powerful and
remarkably simple method for demonstrating that the probability that an inte-
ger has an even number of prime factors (multiplicity included) is equal to the
probability that it has an odd number of prime factors. This finding is closely
connected with the Riemann Hypothesis (RH) and is necessary for its proof.
We use the Liouville function [2], λ(n), defined by λ(n) = (−1)Ω(n) , where
Ω(n) is the total number of prime numbers in the factorization of n, including
the multiplicity of the primes. By definition, λ(1) = 1. The result we prove
states that if n is an arbitrary integer, then

P rob[λ(n) = +1] = P rob[λ(n) = −1] = 1/2.

To demonstrate this result, we recursively construct infinite sets of positive

integers labeled by Ij , where j runs through the prime numbers: 2, 3, 5, 7, ...
First consider the set I2 defined as

I2 = {2k , k = 0, 1, 2, 3, ...}.

The integers in each of the pairs in this set (1, 2), (22 , 23 ), (24 , 25 ), ... are ‘twins’
in the sense that they have λs of opposite signs:

λ(22k ) = −λ(22k+1 ), k = 0, 1, 2, 3, ...

An ordered countably infinite set will be equipartitioned if its members in order

are alternately placed in two different sub-sets. So we conclude that there are
equal proportions of integers with λ’s of +1 and -1 in I2 .

Page 195
 Next, we introduce the prime 3 by defining the set I3 :

I3 = {3l ak , ak ∈ I2 and l = 0, 1, 2, 3, ...}.

This set comprises members of the form

{1, 2, 22 , 23 , ....
3, 3 × 2, 3 × 22 , 3 × 23 , 3 × 23 , ...
32 , 32 × 2, 32 × 22 , 32 × 23 , 32 × 23 , ...
33 , 33 × 2, 33 × 22 , 33 × 23 , 33 × 23 , ...
34 , 34 × 2, 34 × 22 , 34 × 23 , 34 × 23 , ...}
In this set, all the pairs on the first line are twins, as we have seen. In each of
the remaining lines, (32l−1 ak , 32l ak ), l = 1, 2, 3, ... are twins because they have
λs of opposite signs. That is any row which contains a particular number which
contains the factor 32l−1 will have as its ‘twin’ the number just below it which
contains the factor 32l . So we conclude that there are equal proportions of in-
tegers with λ’s of +1 and -1 in I3 .

In a similar manner, we can introduce the primes 5, 7, ...etc. Suppose p is

an arbitrary prime and Ip is the corresponding set recursively constructed as
shown above. If q is the next prime after p, we define

Iq = {q l ak , ∀ak ∈ Ip and l = 0, 1, 2, 3, ...}. (1)

Note that Iq contains all the integers that can be constructed using primes ≤ q.
Therefore, Iq contains all the integers belonging to Im for m < q.

Theorem A: All the integers in the set Iq , q = 2, 3, 5, 7, ...can be paired so that

their λ-values are of the opposite sign.
Proof: Since the claim has already been demonstrated above for q = 2, we
presume that q > 2. Let p be the prime that immediately precedes q, and
suppose that all the integers in the set Ip are twinned.
When l = 0 in Eq. (1) all the integers in Iq are ak , which belong to
Ip are twinned by assumption. When l > 0, for every ak ∈ Ip , the pairs
(q 2l−1 ak , q 2l ak ), l = 1, 2, 3, ... are obviously twins in the sense that

λ(q 2l−1 ak ) = −λ(q 2l ak ), l = 1, 2, 3, ...

Therefore, if the integers in Ip are twinned, so are the integers in Iq . But we

have seen that when p = 2, all the integers in I2 are twinned. Thus the property
claimed in this Theorem follows by induction. QED. 1

Continuing in this manner, we see that all the integers in I∞ can be twinned.
But I∞ comprises the entire set of positive integers. Therefore we have shown
1 There are of course, other methods of ‘twinning’ : If we choose any two numbers

a, b ε Ip such that (a, b) are twins in Ip then the two numbers (q m .a, q m .b) are twins
in Iq for any m ≥ 1 (Remember p is a prime and q is the next higher prime).

Page 196
that the natural numbers have equal proportions with λ’s of +1 and -1. Thus
we have shown:

Theorem B : If n is an arbitrary positive integer,

P rob[λ(n) = +1] = P rob[λ(n) = −1] = 1/2. (2)

The Connection to the Riemann Hypothesis

The condition of equal probabilities proved above is only a necessary condi-
tion. The necessary and sufficient condition for RH to be valid is that for large
N: [3],[4]-[7]:

L(N )
1 = 0 as N → ∞, (3)
N 2 +
for  > 0, where L(N ) is the so called Liouville summatory function defined by
N
X
L(N ) = λ(n). (4)
n=1
The Steps Used to Prove the Riemann Hypothesis
The steps undertaken to prove the Riemann hypothesis are detailed in [5]
(see also its accompanying guide [6]). We briefly summarize the method that
was followed in [5] to establish RH.
1. We demonstrate that the L(N ) series above is a standard random walk,
whose distance travelled in N steps can be represented as:
N
X
X(N ) = xi (5)
i=1

where the xi ’s are independent random numbers with an equal probability

of being either +1 or -1, i.e., essentially “coin-tosses”. It is a well-known
result that the expected value of X(N ), for large N , is
limN →∞ E(|X(N )|) = C0 N 1/2 (6)

2. To do this we have to prove that the λ(n)’s in the L(N ) series are essen-
tially “coin-tosses”, for large n. That is, we have first to show that (i)
their probabilities of being either +1 of -1 are equal, as was proved by
Theorem 3 in [5], and has been shown again, more elegantly, by Theorem
B above.
3. Further, we show (ii) that the λ’s appearing in the natural sequence,
n = 1, 2, 3, ..., are independent of each other for large n. This is done by
two different approaches in Appendices III and IV in [5].

Page 197
 4. We then invoke (towards the end of Section 5 in [5]) Khinchin and Kol-
mogorov’s law of the iterated logarithm [10]-[11] to show that the max-
imum deviation dN , from the one-half power-law expectation, in the ex-
ponent of |X(N )| for any individual random walk tends monotonically to
zero as N → ∞. We have proved above that the L(N ) series is one real-
ization of a random walk. So Khinchin and Kolmogorov’s law also holds
for |L(N )|. Therefore for any chosen  > 0 in (3), the statement for the
validity of RH will be satisfied and the Riemann Hypothesis is proved.

2 Acknowledgements
I thank my brothers Professors Mukesh Eswaran (Univ. of British Columbia,
Canada) and Vinayak Eswaran (IIT, Hyderabad) for their many discussions
and criticisms and for their careful reading of this manuscript. I thank the
management of SNIST for their sustained encouragement and support.

3 References
1. Riemann,B. (1859) in ‘Gessammeltz Werke’, Liepzig,1892,trans., ’On
the number of primes less than a given magnitude’, available in the Edwards
book, op cit., pp299-305.
2. Apostol, T. (1998) Introduction to Analytic Number Theory, Chapter 2,
pp. 37-38, Springer International, Narosa Publishers, New Delhi.
3. Littlewood, J.E. (1912), ‘Quelques consequences de l’hypoth‘ese que la
fonction ζ(s) de Riemann n’a pas de zeros dans le demi-plan R(s) > 1/2,’
Comptes Rendus de l’Acad. des Sciences (Paris), 154, pp. 263–266. Transl.
version,in Edwards op cit.
4. Edwards, H.M. (1974), Riemann’s Zeta Function, Academic Press, New
York.
5. Eswaran, K. (2018), ’A Rigorous Proof of the Riemann Hypothesis from
First Principles’,
https://www.researchgate.net/publication/322697717. Please read the latest
version II, uploaded after the uploading of this paper.
6. Eswaran, K. (2018), ‘A Quick Reading Guide to the Proof of the Riemann
Hypothesis’,
https://www.researchgate.net/publication/324200794
7. Borwein, P., Choi,S., Rooney,B., and Weirathmueller,A., (2006) The Rie-
mann Hypothesis, Springer. Also see slides 28 to 32 in Lecture:
http://www.cecm.sfu.ca/personal/pborwein/SLIDES/RH.pdf
8. Knuth D. (1968), Art of Computer Programming, vol. 2, Chap 3, Addison
Wesley.
9. Chandrasekhar S. (1943), ’Stochastic Problems in Physics and Astron-
omy’, Rev. of Modern Phys., vol.15, no1, pp1-87.

Page 198
 10. Kolmogorov, A. (1929) “Uber das Gesetz des iterierten Logarithmus”,
Mathematische Annalen, 101:126-135.(At the DigitalisierungsZentrum web site)).
Also see: https:en.wikipedia.orgwikiLaw of the iterated logarithm.
11. Khinchine, A. (1924) “Uber einen Satz der Wahrscheinlichkeitsrech-
nung”, Fundamenta Mathematicae 6, pp.9-20.

Page 199
View publication stats